Delete MACS3/Signal/CallPeakUnit.py

[MACS.git] / ChangeLog

2024-10-08  Tao Liu  <vladimir.liu@gmail.com>
        MACS 3.0.3b

        * Features added

        1) We implemented the IO module for reading the fragment files
        usually used in single-cell ATAC-seq experiment
        `Parser.FragParser`. And we implemented a new
        `PairedEndTrack.PETrackII` to store the data in fragment file,
        including the barcodes and counts information. In the `PETrackII`
        class, we are able to extract a subset using a list of barcodes,
        which enables us to call peaks only on a pool (pseudo-bulk) of
        cells.

        2) We extensively rewrote the `pyx` codes into `py` codes. In
        another words, we now apply the 'pure python style' with PEP-484
        type annotations to our previous Cython style codes. So that, the
        source codes can be more compatible to Python programming tools
        such as `flake8`. During rewritting, we cleaned the source codes
        even more, and removed unnecessary dependencies during
        compilation.

        * Bug fixed

        1) Fix issues in big-endian system in `Parser.py` codes. Enable
        big-endian support in `BAM.py` codes for accessig certain
        alignment records that overlap with givin genomic
        coordinates using BAM/BAI files.

        * Doc

        1) Explanation on the filtering criteria on SAM/BAM/BAMPE files.

2024-09-06  Tao Liu  <vladimir.liu@gmail.com>
        MACS 3.0.2

        * Features added

        1) Introduce a new emission model for the `hmmratac` function. Now
        users can choose the simpler Poisson emission `--hmm-type poisson`
        instead of the default Gaussian emission. As a consequence, the
        saved HMM model file in json will include the hmm-type information
        as well. Note that in order to be compatible with the HMM model
        file from previous version, if there is no hmm-type information in
        the model file, the hmm-type will be assigned as gaussian. #635

        2) `hmmratac` now output narrowPeak format output. The summit
        position and the peak score columns reported in the narrowPeak
        output represents the position with highest foldchange value
        (pileup vs average background).

        3) Add `--cutoff-analysis-steps` and `--cutoff-analysis-max` for
        `callpeak`, `bdgpeakcall`, and `hmmratac` so that we can
        have finer resolution of the cutoff analysis report. #636  #642

        4) Reduce memory usage of `hmmratac` during decoding step, by
        writing decoding results to a temporary file on disk (file
        location depends on the environmental TEMP setting), then loading
        it back while identifying state pathes. This change will decrease
        the memory usage dramatically. #628 #640

        5) Fix instructions for preparing narrowPeak files for uploading
        to UCSC browser, with the `--trackline` option in `callpeak`. #653

        6) For gappedPeak output, set thickStart and thickEnd columns as
        0, according to UCSC definition.

        * Bugs fixed

        1) Use `-O3` instead of `-Ofast` for compatibility. #637

        * Documentation

        1) Update instruction to install macs3 through conda/bioconda

        2) Reorganize MACS3 docs and publish through
        https://macs3-project.github.io/MACS

        3) Description on various file formats used in MACS3.

2024-02-19  Tao Liu  <vladimir.liu@gmail.com>
        MACS 3.0.1

        * Bugs fixed

        1) Fixed a bug that the `hmmatac` can't correctly save the
        digested signal files. #605 #611

        2) Applied a patch to remove cython requirement from the installed
        system. (it's needed for building the package). #606 #612

        3) Relax the testing script while comparing the peaks called from
        current codes and the standard peaks. To implement this, we added
        'intersection' function to 'Regions' class to find the
        intersecting regions of two Regions object (similar to PeakIO but
        only recording chromosome, start and end positions). And we
        updated the unit test 'test_Region.py' then implemented a script
        'jaccard.py' to compute the Jaccard Index of two peak files. If
        the JI > 0.99 we would think the peaks called and the standard
        peaks are similar. This is to avoid the problem caused by
        different Numpy/SciPy/sci-kit learn libraries, when certain peak
        coordinates may have 10bps difference. #615 #619

        4) Due to the changes in scikit-learn 1.3.0:
        https://scikit-learn.org/1.3/whats_new/v1.3.html: The way hmmlearn
        0.3 uses Kmeans will end up with inconsistent results between
        sklearn <1.3 and sklearn >=1.3. Therefore, we patched the class
        hmm.GaussianHMM and adjusted the standard output from `hmmratac`
        subcommand. The change is based on
        https://github.com/hmmlearn/hmmlearn/pull/545. The idea is to do
        the random seeding of KMeans 10 times. Now the `hmmratac` results
        should be more consistent (at least JI>0.99). #615 #620

        * Other

        1) We added some dependencies to MACS3. `hmmratc` subcommand needs
        `hmmlearn` library, `hmmlearn` needs `scikit-learn` and
        `scikit-learn` needs `scipy`. Since major releases have happened
        for both`scipy` and `scikit-learn`, we have to set specific
        version requirements for them in order to make sure the output
        results from `hmmratac` are consistent.

        2) We updated our documentation website using
        Sphinx. https://macs3-project.github.io/MACS/

2023-11-15  Tao Liu  <vladimir.liu@gmail.com>
        MACS 3.0.0

        1) Call variants in peak regions directly from BAM files. The
        function was originally developed under code name SAPPER. Now
        SAPPER has been merged into MACS as the `callvar` command. It can
        be used to call SNVs and small INDELs directly from alignment
        files for ChIP-seq or ATAC-seq. We call `fermi-lite` to assemble
        the DNA sequence at the enriched genomic regions (binding sites or
        accessible DNA) and to refine the alignment when necessary. We
        added `simde` as a submodule in order to support fermi-lite
        library under non-x64 architectures.

        2) HMMRATAC module is added as subcommand `hmmratac`. HMMRATAC is
        a dedicated software to analyze ATAC-seq data. The basic idea
        behind HMMRATAC is to digest ATAC-seq data according to the
        fragment length of read pairs into four signal tracks: short
        fragments, mono-nucleosomal fragments, di-nucleosomal fragments
        and tri-nucleosomal fragments. Then integrate the four tracks
        again using Hidden Markov Model to consider three hidden states:
        open region, nucleosomal region, and background region. The
        orginal paper was published in 2019 written in JAVA, by Evan
        Tarbell. We implemented it in Python/Cython and optimize the whole
        process using existing MACS functions and hmmlearn. Now it can run
        much faster than the original JAVA version. Note: evaluation of
        the peak calling results is still underway.

        3) Speed/memory optimization.  Use the cykhash to replace python
        dictionary. Use buffer (10MB) to read and parse input file (not
        available for BAM file parser). And many optimization tweaks. We
        added memory monitoring to the runtime messages.

        4) R wrappers for MACS -- MACSr for bioconductor.

        5) Code cleanup. Reorganize source codes. 

        6) Unit testing. 

        7) Switch to Github Action for CI, support multi-arch testing
        including x64, armv7, aarch64, s390x and ppc64le. We also test on
        Mac OS 12.

        8) MACS tag-shifting model has been refined. Now it will use a
        naive peak calling approach to find ALL possible paired peaks at +
        and - strand, then use all of them to calculate the
        cross-correlation. (a related bug has been fix
        [#442](https://github.com/macs3-project/MACS/issues/442))

        9) BAI index and random access to BAM file now is
        supported. [#449](https://github.com/macs3-project/MACS/issues/449).

        10) Support of Python > 3.10
        [#498](https://github.com/macs3-project/MACS/issues/498)

        11) The effective genome size parameters have been updated
        according to
        deeptools. [#508](https://github.com/macs3-project/MACS/issues/508)

        12) Multiple updates regarding dependencies, anaconda built, CI/CD
        process.

        13) Cython 3 is supported.

        14) Documentations for each subcommand can be found under /docs

        *Other*

        1) Missing header line while no peaks can be called
        [#501](https://github.com/macs3-project/MACS/issues/501)
        [#502](https://github.com/macs3-project/MACS/issues/502)

        2) Note: different numpy, scipy, sklearn may give slightly
        different results for hmmratac results. The current standard
        results for automated testing in `/test` directory are from Numpy
        1.25.1, Scipy 1.11.1, and sklearn 1.3.0.

2020-04-11  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.2.7.1

        * hotfix:

        Add 'wheel' and 'pip' to pyproject.toml so that `pip install` can
        work.

2020-04-10  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.2.7

        * Bugs fixed

        1) MACS2 has been tested on multiple architectures to make sure it
        can successfully generate consistent results. Currently the
        supported architectures are: AMD64, ARM64, i386, PPC64LE, and
        S390X. Thanks to @mr-c, @junaruga, and @tillea! Related to issue
        #340, #349, #351, and #359; to PR #348, #350, #360, #361, #367,
        and #370. The lesson is that if the project is built on Cython and
        is aimed at memory efficiency, we should specifically define all
        int/float types in pyx files such as int8_t or uint32_t using
        either libc or numpy (c version) instead of relying on Cython
        types such as short, long, double.

        2) MACS2 setup script will check numpy and install numpy if
        necessary. PR #378, issue #364

        3) `bdgbroadcall` command will correctly add the score column (5th
        column). The score (5th) column contains 10 times of the average
        score in the broad region. PR #373, issue #362

        4) The missing test on `bdgopt` subcommand has been added. PR #363

        5) The obsolete option `--ratio` from `callpeak` subcommand has
        been removed. PR #369, issue #366

        6) Fixed the incorrect description in README on the 'maximum
        length of broad region is 4 times of d' to 'maximum gap for
        merging broad regions is 4 times of tag size by default'. PR #380,
        issue #365.

        * Other

        1) CODE OF CONDUCT document has been added to MACS2 github
        repository. PR #358

2019-12-12  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.2.6

        * New Features

        1) Speed up MACS2. Some programming tricks and code cleanup. The
        filter_dup function replaces separate_dups. The later one was
        implemented for potentially putting back duplicate reads in
        certain downstream analysis. However such analysis hasn't been
        implemented. Optimize the speed of writing bedGraph
        files. Optimize BAM and BAMPE parsing with pointer casting instead
        of python unpack.

        2) The comment lines in the headers of BED or SAM files will be
        correctly skipped. However, MACS2 won't check comment lines in the
        middle of the file.

        * Bugs fixed

        1) Cutoff-analysis in callpeak command. #341

        2) Issues related to SAMParser and three ELAND Parsers are
        fixed. #347

        * Other

        1) cmdlinetest script in test/ folder has been updated to: 1. test
        cutoff-analysis with callpeak cmd; 2. output the 2 lines before
        and after the error or warning message during tests; 3. output
        only the first 10 lines if the difference between test result and
        standard result can be found; 4. prockreport monitor CPU time and
        memory usage in 1 sec interval -- a bit more accurate.

        2) Python3.5 support is removed. Now MACS2 requires Python>=3.6.

2019-10-31  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.2.5 (Py3 speed up)

        * Features added

        1) *Github code only and Not included in MACS2 release* New
        testing data for performance test. An subsampled ENCODE2 CTCF
        ChIP-seq dataset, including 5million ChIP reads and 5 million
        control reads, has been included in the test folder for testing
        CPU and memory usage (i.e. 5M test). Several related scripts ,
        including `prockreport` for output cpu memory usage, `pyprofile`
        and `pyprofile_stat` for debuging and profiling MACS2 codes, have
        been included.

        2) Speed up pvalue-qvalue checkup (pqtable checkup) #335 #338.
        The old hashtable.pyx implementation copied from Pandas (very old
        version) doesn't work well in Python3+Cython. It slows down the
        pqtable checkup using the identical Cython codes as in
        v2.1.4. While running 5M test, the `__getitem__` function in the
        hashtable.pyx took 3.5s with 37,382,037 calls in MACS2 v2.1.4, but
        148.6s with the same number of calls in MACS2 v2.2.4. As a
        consequence, the standard python dictionary implementation has
        replaced hashtable.pyx for pqtable checkup. Now MACS2 runs a bit
        faster than py2 version, but uses a bit more memory. In general,
        v2.2.5 can finish 5M reads test in 20% less time than MACS2
        v2.1.4, but use 15% more memory.

        * Bug fixed

        1) More Python3 related fixes, e.g. the return value of keys from
        py3 dict. #333 #337


2019-10-01  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.2.4 (Python3)

        * Features added

        1) First Python3 version MACS2 released.

        2) Version number 2.2.X will be used for MACS2 in Python3, in
        parallel to 2.1.X.

        3) More comprehensive test.sh script to check the consistency of
        results from Python2 version and Python3 version.

        4) Simplify setup.py script since the newest version transparently
        supports cython. And when cython is not installed by the user,
        setup.py can still compile using only C codes.

        5) Fix Signal.pyx to use np.array instead of np.mat.

2019-09-30  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.4

        * Features added

        Github Actions is used together with Travis CI for testing and
        deployment.

        * Bugs fixed

        PR #322:

        1) #318 Random score in bdgdiff output. It turns out the sum_v is
        not initialized as 0 before adding. Potential bugs are fixed in
        other functions in ScoreTrack and CallPeakUnit codes.

        2) #321 Cython dependency in setup.py script is removed. And place
        'cythonzie' call to the correct position.

        3) A typo is fixed in Github Actions script.

2019-09-19  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.3.3

        * Features added

        1) Support Docker auto-deploy. PR #309

        2) Support Travis CI auto-testing, update unit-testing
        scripts, and enable subcommand testing on small datasets.

        3) Update README documents. #297 PR #306

        4) `cmbreps` supports more than 2 replicates. Merged from PR #304
        @Maarten-vd-Sande and PR #307 (our own chi-sq test code)

        5) `--d-min` option is added in `callpeak` and `predictd`, to
        exclude predictions of fragment size smaller than the given
        value. Merged from PR #267 @shouldsee.

        6) `--buffer-size` option is added in `predictd`, `filterdup`,
        `pileup` and `refinepeak` subcommands. Users can use this option
        to decrease memory usage while there are a large number of contigs
        in the data. Also, now `callpeak`, `predictd`, `filterdup`,
        `pileup` and `refinepeak` will suggest users to tweak
        `--buffer-size` while catching a MemoryError. #313 PR #314

        * Bugs fixed

        1) #265 Fixed a bug where the pseudocount hasn't been applied
        while calculating p-value score in ScoreTrack object.

        2) Fixed bdgbroadcall so that it will report those broad peaks
        without strong peak inside, a consistent behavior as `callpeak
        --broad`.

        3) Rename COPYING to LICENSE.

2018-10-17  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.2

        * New features

        1) Added missing BEDPE support. And enable the support for BAMPE
        and BEDPE formats in 'pileup', 'filterdup' and 'randsample'
        subcommands. When format is BAMPE or BEDPE, The 'pileup' command
        will pile up the whole fragment defined by mapping locations of
        the left end and right end of each read pair. Thank @purcaro

        2) Added options to callpeak command for tweaking max-gap and
        min-len during peak calling. Thank @jsh58!

        3) The callpeak option "--to-large" option is replaced with
        "--scale-to large".

        4) The randsample option "-t" has been replaced with "-i".

        * Bug fixes

        1) Fixed memory issue related to #122 and #146

        2) Fixed a bug caused by a typo. Related to #249, Thank @shengqh

        3) Fixed a bug while setting commandline qvalue cutoff.

        4) Better describe the 5th column of narrowPeak. Thank @alexbarrera

        5) Fixed the calculation of average fragment length for paired-end
        data. Thank @jsh58

        6) Fixed bugs caused by khash while computing p/q-value and log
        likelihood ratios. Thank @jsh58

        7) More spelling tweaks in source code. Thank @mr-c

2016-03-09  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.1 20160309

        * Retire the tag:rc.

        * Fixed spelling. Merged pull request #120. Thank @mr-c!

        * Change filtering criteria for reading BAM/SAM files

        Related to callpeak and filterdup commands. Now the
        reads/alignments flagged with 1028 or 'PCR/Optical duplicate' will
        still be read although MACS2 may decide them as duplicates
        later. Related to old issue #33. Sorry I forgot to address it for
        years!

2016-02-26  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.1 20160226 (tag:rc Zhengyue)

        * Bug fixes

        1) Now "-Ofast" has been replaced by "-O3 --ffast-math", because
        the former option is not supported by older GCC. Related to issues
        #91, #109.

        2) Issue #108 is fixed. If no peak can be found in a chromosome,
        the PeakIO won't throw an error.

        * New features

        1) callpeak

        a) A more flexible format, BEDPE, is supported. Now users can
        define the left and right position of the ChIPed fragment, and
        MACS2 will skip model building and directly pileup the
        fragments. Related to issue #112.

        b) The 'tempdir' can be specified, to save cached pileup
        tracks. Originially, the temporary files were stored in
        /tmp. Thank @daler! Related to issues #97 and #105.

        2) bdgopt

        New operations are added, to calculate the maximum or minimum value between
        values in BEDGRAPH and given value.

        3) bdgcmp

        New method is added, to calculate the maximum value between values
        defined in two BEDGRAPH files.

2015-12-22  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.0 20151222 (tag:rc Dongzhi)

        * Bug fixes

        1) Fix a bug while dealing with some chromosomes only containing
        one read (pair). The size of dup_plus/dup_minus arrays after
        filtering dups should +1.

        2) Fix a bug related to the broad peak calling function in
        previous versions. The gaps were miscalculated, so segmented weak
        broad calls may be reported, and sometimes you would see peaks
        with lower than cutoff values in the output files.

        3) "Potentially" Fixed issue #105 on temporary cache files, need
        further followup.


2015-07-31  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.0 20150731 (tag:rc)

        * Bug fixes

        1) Fixed issue #76: information about broad/narrow cutoff will be
        correctly displayed.

        2) Fixed issue #79: bdgopt extparam option is fixed.

        3) Fixed issue #87: reference to cProb has been fixed as 'Prob'
        for filterdup command.

        4) Fixed issue #78, #88 and similar issue reported in MACS google
        group: MACS2 now can correctly deal with multiple alignment files
        for -t or -c. The 'finalize' function will be correctly
        called. Multiple files option is enabled for filterdup,
        randsample, predictd, pileup and refinepeak commands.

        5) A related issue to #88, when BAMPE mode is used, PE pairs will
        be sorted by leftmost then rightmost ends. 

        6) Fixed issue #86: A wrong use of 'ndarray' to create Numpy
        array. This will cause 'callpeak --nolambda' hang forever while
        calculating pvalues and qvalues.

2015-04-20  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.0 20150420 (tag:rc)

        * New commands

        1) bdgopt: some convenient functions to modify bedGraph files.

        2) cmbreps: Combine scores from two replicates. Including three
        methods: 1. take the maximum; 2. take the average; 3. use Fisher's
        method to combine two p-value scores. After that, user can use
        bdgpeakcall to call peaks on combined scores.

        * New features

        1) callpeak and bdgpeakcall now can try to analyze the
        relationship between p-values and number/length of peaks then
        generate a summary to help users decide an appropriate cutoff.

        2) callpeak now can accept fold-enrichment cutoff as a filter for
        final peak calls.

        * Performance

        Now MACS2 runs about 3X as fast as previous version. Trade
        clean python codes for speed... Now while processing 50M ChIP vs
        50M control, it will take only 10 minutes.

        * Bug fixes

        1) Sampling function in BAMPE mode.

        2) Callpeak while there are >= 2 input files for -t or -c.

        3) While reading BAM/SAM, those secondary or supplementary
        alignments will be correctly skipped.

        4) Fixed issue #33: Explanation is added to callpeak --keep-dup
        option that MACS2 will discard those SAM/BAM alignments with bit
        1024 no matter how --keep-dup is set.

        5) Fixed issue #49: setuptools is used intead of distutils

        6) Fixed issue #51: fix the problem when using --trackline
        argument when control file is absent.

        7) Fixed issue #53: Use Use SAM/BAM CIGAR to find the 5' end of
        read mapped to minus strand. Previous implementation will find
        incorrect 5' end if there is indel in alignment.

        8) Fixed issue #56: An incorrect sorting method used for BAMPE
        mode which will cause incorrect filtering of duplicated reads. Now
        fixed.

        9) Issue #63: Merged from jayhesselberth@github, extsize now can
        be 1.

        10) Issue #71: Merged from aertslab@github, close file descriptor
        after creating them with mkstemp().

2014-06-16  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.1.0 20140616 (tag:rc)

        * callpeak module

        "--ratio" is added to manually assign the scaling factor of ChIP
        vs control, e.g. from NCIS. Thank Colin D and Dietmar Rieder for
        implementing the patch file!

        "--shift" is added to move cutting ends (5' end of reads) around,
        in order to process DNAse-Seq data, e.g., use "--shift -100
        --extsize 200" to get 200bps fragments around 5' ends. For general
        ChIP-Seq data analysis, this option should be always set as
        0. Thank Xi Chen and Anshul Kundaje for the discussions in user
        group!

        ** Do not output negative fragment size from cross-correlation
        analysis. Thank Alvin Qin for the feedback!

        ** --half-ext and --control-shift are removed. For complex read
        shifting and extending, combine '--shift' and '--extsize'
        options. For comparing two conditions, use 'bdgdiff' module
        instead.

        ** a bug is fixed to output the last pileup value in bdg file
        correctly.

        * filterdup

        A 'dry-run' option is added to only output numbers, including the
        number of allowed duplicates, the total number of reads before and
        after filtering duplicates and the estimated duplication
        rate. Thank John Urban for the suggestion!


2013-12-16  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20131216 (tag:alpha)

        bug fixes and tweaks

        * We changed license from Artistic License to 3-clauses BSD license.

        Yes. Simpler the better.

        * Process paired-end data with "-f BAMPE" without control

        * GappedPeak output for --broad option has been fixed again to be
        consistent with official UCSC format. We add 1bp pseudo-block to
        left and/or right of broad region when necessary, so that you can
        virtualize the regions without strong enrichment inside
        successfully. In downstream analysis except for virtualization,
        you may need to remove all 1bps blocks from gappedPeak file.

        * diffpeak subcommand is temporarily disabled. Till we
        re-implement it.

2013-10-28  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20131028 (tag:alpha)

        * callpeak --call-summits improvement

        The smoothing window length has been fixed as fragment length
        instead of short read length. The larger smoothing window will
        grant better smoothing results and better sub-peak summits
        detection.

        * --outdir and --ofile options for almost all commands

        Thank Björn Grüning for initially implementing these options!
        Now, MACS2 will save results into a specified
        directory by '--outdir' option, and/or save result into a
        specified file by '--ofile' option. Note, in case '--ofile' is
        available for a subcommand, '-o' now has been adjusted to be the
        same as '--ofile' instead of '--o-prefix'.

        Here is the list of changes. For more detail, use 'macs2 xxx -h'
        for each subcommand:

        ** callpeak: --outdir
        ** diffpeak: Not implemented
        ** bdgpeakcall: --outdir and --ofile
        ** bdgbroadcall: --outdir and --ofile
        ** bdgcmp: --outdir and --ofile. While --ofile is used, the number
        and the order of arguments for --ofile must be the same as for -m.
        ** bdgdiff: --outdir and --ofile
        ** filterdup: --outdir
        ** pileup: --outdir
        ** randsample: --outdir
        ** refinepeak: --outdir and --ofile


2013-09-15  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130915 (tag:alpha)

        * callpeak Added a new option --buffer-size

        This option is to tweak a previously hidden parameter that
        controls the steps to increase array size for storing alignment
        information. While in some rare cases, the number of
        chromosomes/contigs/scaffolds is huge, the original default
        setting will cause a huge memory waste. In these cases, we
        recommend to decrease --buffer-size (e.g., 1000) to save memory,
        although the decrease will slow process to read alignment files.

        * an optimization to speed up pvalue-qvalue statistics

        Previously, it took a hour to prepare p-q-table for 65M vs 65M
        human TF library, and now it will take 10 minutes. It was due to a
        single line of code to get a value from a numpy array ...

        * fixed logLR bugs.

2013-07-31  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130731 (tag:alpha)

        * callpeak --call-summits

        Fix bugs causing callpeak --call-summits option generating extra
        number of peaks and inconsistent peak boundaries comparing to
        default option. Thank Ben Levinson!

        * bdgcmp output

        Fix bugs causing bdgcmp output logLR all in positive values. Now
        'depletion' can be correctly represented as negative values.

        * bdgdiff

        Fix the behavior of bdgdiff module. Now it can take four
        bedGraph files, then use logLR as cutoff to call differential
        regions. Check command line of bdgdiff for detail. 

2013-07-13  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130713 (tag:alpha)

        * fix bugs while output broadPeak and gappedPeak.

        Note. Those weak broad regions without any strong enrichment
        regions inside won't be saved in gappedPeak file.

        * bdgcmp -T and -C are merged into -S and description is updated.

        Now, you can use it to override SPMR values in your input for
        bdgcmp. To use SPMR (from 'callpeak --SPMR -B') while calculating
        statistics will cause weird results ( in most cases, lower
        significancy), and won't be consistent with MACS2 callpeak
        behavior. So if you have SPMR bedGraphs, input the smaller/larger
        sample size in MILLION according to 'callpeak --to-large' option.

2013-07-10  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130710 (tag:alpha)

        * fix BED style output format of callpeak module:

        1) without --broad: narrowPeak (BED6+4) and BED for summit will be
        the output. Old BED format file won't be saved.

        2) with --broad: broadPeak (BED6+3) for broad region and
        gappedPeak (BED12+3) for chained enriched regions will be the
        output. Old BED format, narrowPeak format, summit file won't be
        saved.

        * bdgcmp now can accept list of methods to calculate scores. So
        you can run it once to generate multiple types of scores. Thank
        Jon Urban for this suggestion!

        * C codes are re-generated through Cython 0.19.1.

2013-05-21  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130520 (tag:alpha)

        * broad peak calling modules are modified in order to report all
        relexed regions even there is no strong enrichment inside.

2013-05-01  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130501 (tag:alpha)

        * Memory usage is decreased to about 1/4-1/5 of previous usage
        Now, the internal data structure and algorithm are both
        re-organized, so that intermediate data wouldn't be saved in
        memory. Intead they will be calculated on the fly. New MACS2 will
        spend longer time (1.5 to 2 times) however it will use less memory
        so can be more usable on small mem servers.

        * --seed option is added to callpeak and randsample commands
        Thank Mathieu Gineste for this suggestion!

2013-03-05  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130306 (tag:alpha)

        * diffpeak module New module to detect differential binding sites
        with more statistics.

        * Introduced --refine-peaks
        Calculates reads balancing to refine peak summits

        * Ouput file names prefix
        Correct encodePeak to narrowPeak, broadPeak to bed12. 

2012-09-13 Benjamin Schiller <benjamin.schiller@ucsf.edu>,  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.10 (tag:alpha not released)

        * Introduced BAMPEParser
        Reads PE data directly, requires bedtools for now

        * Introduced --call-summits
        Uses signal processing methods to call overlapping peaks

        * Added --no-trackline
        By default, files have descriptive tracklines now

        * new refinepeak command (experimental)
        This new function will use a similar method in SPP (wtd), to
        analyze raw tag distribution in peak region, then redefine the
        peak summit where plus and minus tags are evenly distributed
        around.

        * Changes to output *
        cPeakDetect.pyx has full support for new print/write methods and
        --call-peaks, BAMPEParser, and use of paired-end data

        * Parser optimization

        cParser.pyx is rewritten to use io.BufferedReader to speed
        up. Speed is doubled.

        Code is reorganized -- most of functions are inherited from
        GenericParser class.

        * Use cross-correlation to calculate fragment size

        First, all pairs will be used in prediction for fragment
        size. Previously, only no more than 1000 pairs are used. Second,
        cross-correlation is used to find the best phase difference
        between + and - tag pileups.

        * Speed up p-value and q-value calculation

        This part is ten times faster now. I am using a dictionary to
        cache p-value results from Poisson CDF function. A bit more memory
        will be used to increase speed. I hope this dictionary would not
        explode since the possible pairs of ChIP signal and control lambda
        are hugely redundant. Also, I rewrited part of q-value
        calculation.

        * Speed up peak detection

        This part is about hundred of times faster now.  Optimizations
        include using Numpy functions as much as possible, and making loop
        body as small as possible.

        * Post-processing on differential calls

        After macs2diff finds differential binding sites between two
        conditions, it will try to annotate the peak calls from one of two
        conditions, describe the changes ...

        * Fragment size prediction in macs2diff

        Now by default, macs2diff will try to use the average fragment
        size from both condition 1 and condition 2 for tag extension and
        peak calling. Previously, by default, it will use different sizes
        unless --nomodel is specified.

        Technically, I separate model building processes out. So macs2diff
        will build fragment sizes for condition 1 and 2 in parallel (2
        processes maximum), then perform 4-way comparisons in parallel (4
        processes maximum).

        * Diff score

        Combine two p/qscore tracks together. At regions where condition 1
        is higher than condition 2, score would be positive, otherwise,
        negative.

        * SAMParser and BAMParser

        Bug fixed for paired-end sequencing data.

        * BedGraph.pyx

        Fixed a bug while calling peaks from BedGraph file. It previously
        mistakenly output same peaks multiple times at the end of
        chromosome.

2011-11-2  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.9 (tag:alpha)

        * Auto fixation on predicted d is turned off by default!

        Previous --off-auto is now default. MACS will not automatically
        fix d less than 2 times of tag size according to
        --shiftsize. While tag size is getting longer nowadays, it would
        be easier to have d less than 2 times of tag size, however d may
        still be meaningful and useful. Please judge it using your own
        wisdom.

        * Scaling issue

        Now, the default scaling while treatment and input are unbalanced
        has been adjusted. By default, larger sample will be scaled down
        linearly to match the smaller sample. In this way, background
        noise will be reduced more than real signals, so we expect to have
        more specific results than the other way around (i.e. --to-large
        is set).

        Also, an alternative option to randomly sample larger data
        (--down-sample) is provided to replace default linear
        scaling. However, this option will cause results irresproducible,
        so be careful. 

        * randsample script

        A new script 'randsample'  is added, which can randomly sample
        certain percentage or number of tags.

        * Peak summit

        Now, MACS will decide peak summits according to pileup height
        instead of qvalue scores. In this way, the summit may be more
        accurate. 

        * Diff score

        MACS calculate qvalue scores as differential scores. When compare
        two conditions (saying A and B), the maximum qscore for comparing
        A to B -- maxqscore_a2b, and for comparing B to A --maxqscore_b2a
        will be computed. If maxqscore_a2b is bigger, the diff score is
        +maxqscore_a2b, otherwise, diff score is -1*maxqscore_b2a.

2011-09-15  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.8 (tag:alpha)

        * bin/macs2, bin/bdgbroadcall, MACS2/IO/cScoreTrack.pyx, MACS2/IO/cBedGraph.pyx

        New script bdgbroadcall and the extra option '--broad' for macs2
        script, can be used to call broad regions with a loose cutoff to
        link nearby significant regions. The output is represented as
        BED12 format.

        * MACS2/IO/cScoreTrack.pyx

        Fix q-value calculation to generate forcefully monotonic values.

        * bin/eland*2bed, bin/sam2bed and bin/filterdup

        They are combined to one more powerful script called
        "filterdup". The script filterdup can filter duplicated reads
        according to sequencing depth and genome size. The script can also
        convert any format supported by MACS to BED format.

2011-08-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.7 (tag:alpha)

        * bin/macsdiff renamed to bin/bdgdiff

        Now this script will work as a low-level finetuning tool as bdgcmp
        and bdgpeakcall.

        * bin/macs2diff

        A new script to take treatment and control files from two
        condition, calculate fragment size, use local poisson to get
        pvalues and BH process to get qvalues, then combine 4-ways result
        to call differential sites.

        This script can use upto 4 cpus to speed up 4-ways calculation. (
        I am trying multiprocessing in python. )

        * MACS2/Constants.py, MACS2/IO/cBedGraph.pyx,
        MACS2/IO/cScoreTrack.pyx, MACS2/OptValidator.py,
        MACS2/PeakModel.py, MACS2/cPeakDetect.pyx

        All above files are modified for the new macs2diff script.

        * bin/macs2, bin/macs2diff, MACS2/OptValidator.py

        Now q-value 0.01 is the default cutoff. If -p is specified,
        p-value cutoff will be used instead.

2011-07-25  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.6 (tag:alpha)

        * bin/macsdiff

        A script to call differential regions. A naive way is introduced
        to find the regions where:

        1. signal from condition 1 is larger than input 1 and condition 2 --
        unique region in condition 1;
        2. signal from condition 2 is larger than input 2 and condition 1
        -- unique region in condition 2;
        3. signal from condition 1 is larger than input 1, signal from
        condition 2 is larger than input 2, however either signal from
        condition 1 or 2 is not larger than the other.

        Here 'larger' means the pvalue or qvalue from a Poisson test is
        under certain cutoff.

        (I will make another script to wrap up mulitple scripts for
        differential calling)

2011-07-07  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.5 (tag:alpha)

        * bin/macs2, MACS2/cPeakDetect.py, MACS2/IO/cScoreTrack.pyx,
        MACS2/IO/cPeakIO.pyx

        Use hash to store peak information. Add back the feature to deal
        with data without control.

        Fix bug which incorrectly allows small peaks at the end of
        chromosomes.

        * bin/bdgpeakcall, bin/bdgcmp

        Fix bugs. bdgpeakcall can output encodePeak format.

2011-06-22  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.4 (tag:alpha)

        * cPeakDetect.py

        Fix a bug, correctly assign lambda_bg while --to-small is
        set. Thanks Junya Seo!

        Add rank and num of bp columns to pvalue-qvalue table.

        * cScoreTrack.py

        Fix bugs to correctly deal with peakless chromosomes. Thanks
        Vaibhav Jain!

        Use AFDR for independent tests instead.

        * encodePeak

        Now MACS can output peak coordinates together with pvalue, qvalue,
        summit positions in a single encodePeak format (designed for
        ENCODE project) file. This file can be loaded to UCSC
        browser. Definition of some specific columns are: 5th:
        int(-log10pvalue*10), 7th: fold-change, 8th: -log10pvalue, 9th:
        -log10qvalue, 10th: relative summit position to peak start.


2011-06-19  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.3 (tag:alpha)

        * Rich output with qvalue, fold enrichment, and pileup height

        Calculate q-values using a refined Benjamini–Hochberg–Yekutieli
        procedure:

        http://en.wikipedia.org/wiki/False_discovery_rate#Dependent_tests

        Now we have a similiar xls output file as before. The differences
        from previous file are:

        1. Summit now is absolute summit, instead of relative summit
           position;
        2. 'Pileup' is previous 'tag' column. It's the extended fragment
           pileup at the peak summit;
        3. We now use '-log10(pvalue)' instead of '-10log10(pvalue)', so
           5.00 means 1e-5, simple and less confusing.
        4. FDR column becomes '-log10(qvalue)' column.
        5. The pileup, -log10pvalue, fold_enrichment and -log10qvalue are
           the values at the peak summit.

        * Extra output files

        NAME_pqtable.txt contains pvalue and qvalue relationships.

        NAME_treat_pvalue.bdg and NAME_treat_qvalue.bdg store -log10pvalue
        and -log10qvalue scores in BedGraph format. Nearby regions with
        the same value are not merged.

        * Separation of FeatIO.py

        Its content has been divided into cPeakIO.pyx, cBedGraph.pyx, and
        cFixWidthTrack.pyx. A modified bedGraphTrackI class was
        implemented to store pileup, local lambda, pvalue, and qvalue
        alltogether in cScoreTrack.pyx.

        * Experimental option --half-ext

        Suggested by NPS algorithm, I added an experimental option
        --half-ext to let MACS only extends ChIP fragment around its
        middle point for only 1/2 d.

2011-06-12  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.2 (tag:alpha)

        * macs2

        Add an error check to see if there is no common chromosome names
        from treatment file and control file

        * cPeakDetect.pyx, cFeatIO.pyx, cPileup.pyx

        Reduce memory usage by removing deepcopy() calls.

        * Modify README documents and others.

2011-05-19  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS Version 2.0.1 (tag:alpha)

        * cPileup.pyx, cPeakDetect.pyx and peak calling process

        Jie suggested me a brilliant simple method to pileup fragments
        into bedGraph track. It works extremely faster than the previous
        function, i.e, faster than MACS1.3 or MACS1.4. So I can include
        large local lambda calculation in MACSv2 now. Now I generate three
        bedGraphs for d-size local bias, slocal-size and llocal-size local
        bias, and calculate the maximum local bias as local lambda
        bedGraph track.

        Minor: add_loc in bedGraphTrackI now can correctly merge the
        region with its preceding region if their value are the same.

        * macs2

        Add an option to shift control tags before extension. By default,
        control tags will be extended to both sides regardless of strand
        information.

2011-05-17  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS Version 2.0.0 (tag:alpha)

        * Use bedGraph type to store data internally and externally.

        We can have theoretically one-basepair resolution profiles. 10
        times smaller in filesize and even smaller after converting to
        bigWig for visualization.

        * Peak calling process modified. Better peak boundary detection.

        Extend ChIP tag to d, and pileup to have a ChIP bedGraph. Extend
        Control tag to d and 1,000bp, and pileup to two bedGraphs. (1000bp
        one will be averaged to d size) Then calculate the maximum value
        of these two tracks and a global background, to have a
        local-lambda bedGraph.

        Use -10log10poisson_pvalue as scores to generate a score track
        before peak calling.

        A general peak calling based on a score cutoff, min length of peak
        and max gap between nearby peaks.

        * Option changes.

        Wiggle file output is removed. Now we only support bedGraph
        output. The generation of bedGraph is highly recommended since it
        will not cost extra time. In other words, bedGraph generation is
        internally run even you don't want to save bedGraphs on disk, due
        to the peak calling algorithm in MACS v2.

        * cProb.pyx

        We now can calculate poisson pvalue in log space so that the score
        (-10*log10pvalue) will not have a upper limit of 3100 due to
        precision of float number.

        * Cython is adopted to speed up Python code.

2011-02-28  Tao Liu  <taoliu@jimmy.harvard.edu>
        Small fixes

        * Replaced with a newest WigTrackI class and fixed the wignorm script.

2011-02-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0rc2 (Valentine)

        * --single-wig option is renamed to --single-profile

        * BedGraph output with --bdg or -B option.

        The BedGraph output provides 1bp resolution fragment pileup
        profile. File size is smaller than wig file. This option can be
        combined with --single-profile option to produce a bedgraph file
        for the whole genome. This option can also make --space,
        --call-subpeaks invalid.

        * Fix the description of --shiftsize to correctly state that the
        value is 1/2 d (fragment size).

        * Fix a bug in the call to __filter_w_control_tags when control is
        not available.

        * Fix a bug on --to-small option. Now it works as expected.

        * Fix a bug while counting the tags in candidate peak region, an
        extra tag may be included. (Thanks to Jake Biesinger!)

        * Fix the bug for the peaks extended outside of chromosome
        start. If the minus strand tag goes outside of chromosome start
        after extension of d, it will be thrown out.

        * Post-process script for a combined wig file:

        The "wignorm" command can be called after a full run of MACS14 as
        a postprocess. wignorm can calculate the local background from the
        control wig file from MACS14, then use either foldchange,
        -10*log10(pvalue) from possion test, or difference after asinh
        transformation as the score to build a single wig track to
        represent the binding strength. This script will take a
        significant long time to process.

        * --wigextend has been obsoleted.

2010-09-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0rc1 (Starry Sky)

        * Duplicate reads option

        --keep-dup behavior is changed. Now user can specify how many
        reads he/she wants to keep at the same genomic location. 'auto' to
        let MACS decide the number based on binomial distribution, 'all'
        to let MACS keep all reads.

        * pvalue and FDR fixes (Thanks to Prof. Zhiping Weng)

        By default, MACS will now scale the smaller dataset to the bigger
        dataset. For instance, if IP has 10 million reads, and Input has 5
        million, MACS will double the lambda value calculated from Input
        reads while calling BOTH the positive peaks and negative
        peaks. This will address the issue caused by unbalanced numbers of
        reads from IP and Input. If --to-small is turned on, MACS will
        scale the larger dataset to the smaller one. So from now on, if d
        is fixed, then the peaks from a MACS call for A vs B should be
        identical to the negative peaks from a B vs A.

2010-09-01  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0beta (summer wishes)

        * New features

        ** Model building

        The default behavior in the model building step is slightly
        changed. When MACS can't find enough pairs to build model
        (implemented in alpha version) or the modeled fragment length is
        less than 2 times of tag length (implemented in beta version),
        MACS will use 2 times of --shiftsize value as fragment length in
        the later analysis. --off-auto can turn off this default behavior.

        ** Redundant tag filtering

        The IO module is rewritten. The redundant tag filtering process
        becomes simpler and works as promise. The maximum allowed number
        of tags at the exact same location is calculated from the
        sequencing depth and genome size using a binomial distribution,
        for both TREAMENT and CONTROL separately. ( previously only
        TREATMENT is considered ) The exact same location means the same
        coordination and the same strand. Then MACS will only keep at most
        this number of tags at the exact same location in the following
        analysis. An option --keep-dup can let MACS skip the filtering and
        keep all the tags. However this may bring in a lot of sequencing
        bias, so you may get many false positive peaks.

        ** Single wiggle mode

        First thing to mention, this is not the score track that I
        described before. By default, MACS generates wiggle files for
        fragment pileup for every chromosomes separately. When you use
        --single-wig option, MACS will generate a single wiggle file for
        all the chromosomes so you will get a wig.gz for TREATMENT and
        another wig.gz for CONTROL if available.

        ** Sniff -- automatic format detection

        Now, by default or "-f AUTO", MACS will decide the input file
        format automatically. Technically, it will try to read at most
        1000 records for the first 10 non-comment lines. If it succeeds,
        the format is decided. I recommend not to use AUTO and specify the
        right format for your input files, unless you combine different
        formats in a single MACS run.

        * Options changes

        --single-wig and --keep-dup are added. Check previous section in
        ChangeLog for detail.

        -f (--format) AUTO is now the default option.

        --slocal default: 1000
        --llocal default: 10000

        * Bug fixed

        Setup script will stop the installation if python version is not
        python2.6 or python2.7.

        Local lambda calculation has been changed back. MACS will check
        peak_region, slocal( default 1K) and llocal (default 10K) for the
        local bias. The previous 200bps default will cause MACS misses
        some peaks where the input bias is very sharp.

        sam2bed.py script is corrected.

        Relative pos in xls output is fixed.

        Parser for ELAND_export is fixed to pass some of the no match
        lines. And elandexport2bed.py is fixed too. ( however I can't
        guarantee that it works on any eland_export files. )

2010-06-04  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0alpha2 (be smarter)

        * Options changes

        --gsize now provides shortcuts for common genomes, including
        human, mouse, C. elegans and fruitfly.

        --llocal now will be 5000 bps if there is no input file, so that
        local lambda doesn't overkill enriched binding sites.

2010-06-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4alpha (be smarter)
        
        * Options changes

        --tsize option is redesigned. MACS will use the first 10 lines of
        the input to decide the tag size. If user specifies --tsize, it
        will override the auto decided tsize.

        --lambdaset is replaced by --slocal and --llocal which mean the
        small local region and large local region. 

        --bw has no effect on the scan-window size now. It only affects the
        paired-peaks model process. 
        
        * Model building

        During the model building, MACS will pick out the enriched regions
        which are not too high and not too low to build the paired-peak
        model. Default the region is from fold 10 to fold 30. If MACS
        fails to build the model, by default it will use the nomodel
        settings, like shiftsize=100bps, to shift and extend each
        tags. This behavior can be turned off by '--off-auto'.

        * Output files

        An extra file including all the summit positions are saved in
        *_summits.bed file. An option '--call-subpeaks' will invoke
        PeakSplitter developed by Mali Salmon to split wide peaks into
        smaller subpeaks.
        
        * Sniff ( will in beta )

        Automatically recognize the input file format, so use can combine
        different format in one MACS run.

        Not implemented features/TODO:
        
        * Algorithms ( in near future? )

        MACS will try to refine the peak boundaries by calculating the
        scores for every point in the candidate peak regions. The score
        will be the -10*log(10,pvalue) on a local poisson distribution. A
        cutoff specified by users (--pvalue) will be applied to find the
        precise sub-peaks in the original candidate peak region. Peak
        boudaries and peak summits positions will be saved in separate BED
        files.

        * Single wiggle track ( in near future? )

        A single wiggle track will be generated to save the scores within
        candidate peak regions in the 10bps resolution. The wiggle file
        is in fixedStep format.


2009-10-16  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.7.1 (Oktoberfest, bug fixed #1)
        
        * bin/Constants.py

        Fixed typo. FCSTEP -> FESTEP

        * lib/PeakDetect.py

        The 'femax' attribute bug is fixed

2009-10-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.7 (Oktoberfest)
        
        * bin/macs, lib/PeakDetect.py, lib/IO/__init__.py, lib/OptValidator.py

        Enhancements by Peter Chines:

        1. gzip files are supported. 
        2. when --diag is on, user can set the increment and endpoint for
        fold enrichment analysis by setting --fe-step and --fe-max.

        Enhancements by Davide Cittaro:

        1. BAM and SAM formats are supported.
        2. small changes in the header lines of wiggle output.

        Enhancements by Me:
        1. I added --fe-min option;
        2. Bowtie ascii output with suffix ".map" is supported.
        
        Bug fixed:

        1. --nolambda bug is fixed. ( reported by Martin in JHU )
        2. --diag bug is fixed. ( reported by Bogdan Tanasa )
        3. Function to remove suffix '.fa' is fixed. ( reported by Jeff Johnston )
        4. Some "fold change" have been changed to "fold enrichment".

2009-06-10  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.6.1 (default parameter change)
        
        * bin/macs, lib/PeakDetect.py

        "--oldfdr" is removed. The 'oldfdr' behaviour becomes
        default. "--futurefdr" is added which can turn on the 'new' method
        introduced in 1.3.6. By default it's off.

        * lib/PeakDetect.py

        Fixed a bug. p-value is corrected a little bit.
        

2009-05-11  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.6 (Birthday cake)
        
        * bin/macs

        "track name" is added to the header of BED output file.

        Now the default peak detection method is to consider 5k and 10k
        nearby regions in treatment data and peak location, 1k, 5k, and
        10k regions in control data to calculate local bias. The old
        method can be called through '--old' option.

        Information about how many total/unique tags in treatment or
        control will be saved in final .xls output.

        * lib/IO/__init__.py

        ".fa" will be removed from input tag alignment so only the
        chromosome names are kept.

        WigTrackI class is added for Wiggle like data structure. (not used
        now)

        The parser for ELAND multi PET files has been fixed. Now the 5'
        tag position for a pair will be kept, whereas in the previous
        version, the middle points are kept.

        * lib/IO/BinKeeper.py

        BinKeeperI class is inspired by Jim Kent's library for UCSC genome
        browser, which can quickly access certain region for values in a
        large wiggle like data file. (not used now)

        * lib/OptValidator.py

        typo fixed.

        * lib/PeakDetect.py

        Now the default peak detection method is to consider 5k and 10k
        nearby regions in treatment data and peak location, 1k, 5k, and
        10k regions in control data to calculate local bias. The old
        method can be called through '--old' option.

        Two columns have beed added to BED output file. 4th column: peak
        name; 5th column: peak score using -10log(10,pvalue) as score.

        * setup.py

        Add support to build a Mac App through 'setup.py py2app', or a
        Windows executable through 'setup.py py2exe'. You need to install
        py2app or py2exe package in order to use these functions.

2009-02-12  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.5 (local lambda fixed, typo fixed, model figure improved)
        
        * PeakDetect.py

        Now, besides 1k, 5k, 10k, MACS will also consider peak size region
        in control data to calculate local lambda for each peak. Peak
        calling results will be slightly different with previous version,
        beware!

        * OptValidator.py

        Typo fixed, ELANDParser -> ELANDResultParser

        * OutputWriter.py

        Now, modeled d value will be shown on the model figure.

2009-01-06  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.4 (Happy New Year Version, bug fixed, ELAND multi/PET support)
        
        * macs, IO/__init__.py, PeakDetect.py

        Add support for ELAND multi format. Add support for Pair-End
        experiment, in this case, 5'end and 3'end ELAND multi format files
        are required for treatment or control data. See 00README file for
        detail.

        Add wigextend option.

        Add petdist option for Pair-End Tag experiment, which is the best
        distance between 5' and 3' tags.

        * PeakDetect.py

        Fixed a bug which cause the end positions of every peak region
        incorrectly added by 1 bp. ( Thanks Mali Salmon!)

        * OutputWriter.py
        
        Fix bugs while generating wiggle files. The start position of
        wiggle file is set to 1 instead of 0.

        Fix a bug that every 10M bps, signals in the first 'd' range are
        lower than actual. ( Thanks Mali Salmon!)


2008-12-03  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.3 (wiggle bugs fixed)
        
        * OutputWriter.py

        Fix bugs while generating wiggle files. 1. 'span=' is added to
        'variableStep' line; 2. previously, every 10M bps, the coordinates
        were wrongly shifted to the right for 'd' basepairs.

        * macs, PeakDetect.py

        Add an option to save wiggle files on different resolution.
        
2008-10-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.2 (tiny bugs fixed)

        * IO/__init__.py
        
        Fix 65536 -> 65535. ( Thank Joon) 
        
        * Prob.py
        
        Improved for binomial function with extra large number. Imported
        from Cistrome project.
        
        * PeakDetect.py

        If treatment channel misses reads in some chromosome included in
        control channel, or vice versa, MACS will not exit. (Thank Shaun
        Mahony)

        Instead, MACS will fake a tag at position -1 when calling
        treatment peaks vs control, but will ignore the chromosome while
        calling negative peaks.

2008-09-04  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.1 (tiny bugs fixed version)

        * Prob.py
        
        Hyunjin Gene Shin contributed some codes to Prob.py. Now the
        binomial functions can tolerate large and small numbers.
        
        * IO/__init__.py

        Parsers now split lines in BED/ELAND file using any
        whitespaces. 'track' or 'browser' lines will be regarded as
        comment lines. A bug fixed when throwing StrandFormatError. The
        maximum redundant tag number at a single position can be no less
        than 65536.

        
2008-07-15  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3 (naming clarification version)

        * Naming clarification changes according to our manuscript:

        'frag_len' is changed to 'd'.

        'fold_change' is changed to 'fold_enrichment'.
        
        Suggest '--bw' parameter to be determined by users from the real
        sonication size.

        Maximum FDR is 100% in the output file.

        And other clarifications in 00README file and the documents on the
        website.
        
        * IO/__init__.py
        If the redundant tag number at a single position is over 32767,
        just remember 32767, instead of raising an overflow exception.
        
        * setup.py
        fixed a typo.

        * PeakDetect.py
        Bug fixed for diagnosis report.
        

2008-07-10  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.2.2gamma

        * Serious bugs fix: 

        Poisson distribution CDF and inverse CDF functions are
        corrected. They can produce right results even for huge lambda
        now. So that the p-value and FDR values in the final excel sheet
        are corrected.

        IO package now can tolerate some rare cases; ELANDParser in IO
        package is fixed. (Thank Bogdan)

        * Improvement:

        Reverse paired peaks in model are rejected. So there will be no
        negative 'frag_len'. (Thank Bogdan)

        * Features added:
        
        Diagnosis function is completed. Which can output a table file for
        users to estimate their sequencing depth.


2008-06-30  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.2
        
        * Probe.py is added!  

        GSL is totally removed from MACS. Instead, I have implemented the
        CDF and inverse CDF for poisson and binomial distribution purely
        in python.

        * Constants.py is added!

        Organize constants used in MACS in the Constants.py file.

        * All other files are modified!

        Foldchange calculation is modified. Now the foldchange only be
        calculated at the peak summit position instead of the whole peak
        region. The values will be higher and more robust than before.

        Features added:

        1. MACS can save wiggle format files containing the tag number at
        every 10 bp along the genome. Tags are shifted according to our
        model before they are calculated.

        2. Model building and local lambda calculation can be skipped with
        certain options.

        3. A diagnosis report can be generated through '--diag'
        option. This report can help you get an assumption about the
        sequencing saturation. This funtion is only in beta stage.

        4. FDR calculation speed is highly improved.
        
2008-05-28  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.1
        
        * TabIO, PeakModel.py ...
        Bug fixed to let MACS tolerate some cases while there is no tag on
        either plus strand or minus strand.

        * setup.py
        Check the version of python. If the version is lower than 2.4,
        refuse to install with warning.


2013-07-31  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130731 (tag:alpha)

        * callpeak --call-summits

        Fix bugs causing callpeak --call-summits option generating extra
        number of peaks and inconsistent peak boundaries comparing to
        default option. Thank Ben Levinson!

        * bdgcmp output

        Fix bugs causing bdgcmp output logLR all in positive values. Now
        'depletion' can be correctly represented as negative values.

        * bdgdiff

        Fix the behavior of bdgdiff module. Now it can take four
        bedGraph files, then use logLR as cutoff to call differential
        regions. Check command line of bdgdiff for detail. 

2013-07-13  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130713 (tag:alpha)

        * fix bugs while output broadPeak and gappedPeak.

        Note. Those weak broad regions without any strong enrichment
        regions inside won't be saved in gappedPeak file.

        * bdgcmp -T and -C are merged into -S and description is updated.

        Now, you can use it to override SPMR values in your input for
        bdgcmp. To use SPMR (from 'callpeak --SPMR -B') while calculating
        statistics will cause weird results ( in most cases, lower
        significancy), and won't be consistent with MACS2 callpeak
        behavior. So if you have SPMR bedGraphs, input the smaller/larger
        sample size in MILLION according to 'callpeak --to-large' option.

2013-07-10  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130710 (tag:alpha)

        * fix BED style output format of callpeak module:

        1) without --broad: narrowPeak (BED6+4) and BED for summit will be
        the output. Old BED format file won't be saved.

        2) with --broad: broadPeak (BED6+3) for broad region and
        gappedPeak (BED12+3) for chained enriched regions will be the
        output. Old BED format, narrowPeak format, summit file won't be
        saved.

        * bdgcmp now can accept list of methods to calculate scores. So
        you can run it once to generate multiple types of scores. Thank
        Jon Urban for this suggestion!

        * C codes are re-generated through Cython 0.19.1.

2013-05-21  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130520 (tag:alpha)

        * broad peak calling modules are modified in order to report all
        relexed regions even there is no strong enrichment inside.

2013-05-01  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130501 (tag:alpha)

        * Memory usage is decreased to about 1/4-1/5 of previous usage
        Now, the internal data structure and algorithm are both
        re-organized, so that intermediate data wouldn't be saved in
        memory. Intead they will be calculated on the fly. New MACS2 will
        spend longer time (1.5 to 2 times) however it will use less memory
        so can be more usable on small mem servers.

        * --seed option is added to callpeak and randsample commands
        Thank Mathieu Gineste for this suggestion!

2013-03-05  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.10 20130306 (tag:alpha)

        * diffpeak module New module to detect differential binding sites
        with more statistics.

        * Introduced --refine-peaks
        Calculates reads balancing to refine peak summits

        * Ouput file names prefix
        Correct encodePeak to narrowPeak, broadPeak to bed12. 

2012-09-13 Benjamin Schiller <benjamin.schiller@ucsf.edu>,  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.10 (tag:alpha not released)

        * Introduced BAMPEParser
        Reads PE data directly, requires bedtools for now

        * Introduced --call-summits
        Uses signal processing methods to call overlapping peaks

        * Added --no-trackline
        By default, files have descriptive tracklines now

        * new refinepeak command (experimental)
        This new function will use a similar method in SPP (wtd), to
        analyze raw tag distribution in peak region, then redefine the
        peak summit where plus and minus tags are evenly distributed
        around.

        * Changes to output *
        cPeakDetect.pyx has full support for new print/write methods and
        --call-peaks, BAMPEParser, and use of paired-end data

        * Parser optimization

        cParser.pyx is rewritten to use io.BufferedReader to speed
        up. Speed is doubled.

        Code is reorganized -- most of functions are inherited from
        GenericParser class.

        * Use cross-correlation to calculate fragment size

        First, all pairs will be used in prediction for fragment
        size. Previously, only no more than 1000 pairs are used. Second,
        cross-correlation is used to find the best phase difference
        between + and - tag pileups.

        * Speed up p-value and q-value calculation

        This part is ten times faster now. I am using a dictionary to
        cache p-value results from Poisson CDF function. A bit more memory
        will be used to increase speed. I hope this dictionary would not
        explode since the possible pairs of ChIP signal and control lambda
        are hugely redundant. Also, I rewrited part of q-value
        calculation.

        * Speed up peak detection

        This part is about hundred of times faster now.  Optimizations
        include using Numpy functions as much as possible, and making loop
        body as small as possible.

        * Post-processing on differential calls

        After macs2diff finds differential binding sites between two
        conditions, it will try to annotate the peak calls from one of two
        conditions, describe the changes ...

        * Fragment size prediction in macs2diff

        Now by default, macs2diff will try to use the average fragment
        size from both condition 1 and condition 2 for tag extension and
        peak calling. Previously, by default, it will use different sizes
        unless --nomodel is specified.

        Technically, I separate model building processes out. So macs2diff
        will build fragment sizes for condition 1 and 2 in parallel (2
        processes maximum), then perform 4-way comparisons in parallel (4
        processes maximum).

        * Diff score

        Combine two p/qscore tracks together. At regions where condition 1
        is higher than condition 2, score would be positive, otherwise,
        negative.

        * SAMParser and BAMParser

        Bug fixed for paired-end sequencing data.

        * BedGraph.pyx

        Fixed a bug while calling peaks from BedGraph file. It previously
        mistakenly output same peaks multiple times at the end of
        chromosome.

2011-11-2  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.9 (tag:alpha)

        * Auto fixation on predicted d is turned off by default!

        Previous --off-auto is now default. MACS will not automatically
        fix d less than 2 times of tag size according to
        --shiftsize. While tag size is getting longer nowadays, it would
        be easier to have d less than 2 times of tag size, however d may
        still be meaningful and useful. Please judge it using your own
        wisdom.

        * Scaling issue

        Now, the default scaling while treatment and input are unbalanced
        has been adjusted. By default, larger sample will be scaled down
        linearly to match the smaller sample. In this way, background
        noise will be reduced more than real signals, so we expect to have
        more specific results than the other way around (i.e. --to-large
        is set).

        Also, an alternative option to randomly sample larger data
        (--down-sample) is provided to replace default linear
        scaling. However, this option will cause results irresproducible,
        so be careful. 

        * randsample script

        A new script 'randsample'  is added, which can randomly sample
        certain percentage or number of tags.

        * Peak summit

        Now, MACS will decide peak summits according to pileup height
        instead of qvalue scores. In this way, the summit may be more
        accurate. 

        * Diff score

        MACS calculate qvalue scores as differential scores. When compare
        two conditions (saying A and B), the maximum qscore for comparing
        A to B -- maxqscore_a2b, and for comparing B to A --maxqscore_b2a
        will be computed. If maxqscore_a2b is bigger, the diff score is
        +maxqscore_a2b, otherwise, diff score is -1*maxqscore_b2a.

2011-09-15  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.8 (tag:alpha)

        * bin/macs2, bin/bdgbroadcall, MACS2/IO/cScoreTrack.pyx, MACS2/IO/cBedGraph.pyx

        New script bdgbroadcall and the extra option '--broad' for macs2
        script, can be used to call broad regions with a loose cutoff to
        link nearby significant regions. The output is represented as
        BED12 format.

        * MACS2/IO/cScoreTrack.pyx

        Fix q-value calculation to generate forcefully monotonic values.

        * bin/eland*2bed, bin/sam2bed and bin/filterdup

        They are combined to one more powerful script called
        "filterdup". The script filterdup can filter duplicated reads
        according to sequencing depth and genome size. The script can also
        convert any format supported by MACS to BED format.

2011-08-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.7 (tag:alpha)

        * bin/macsdiff renamed to bin/bdgdiff

        Now this script will work as a low-level finetuning tool as bdgcmp
        and bdgpeakcall.

        * bin/macs2diff

        A new script to take treatment and control files from two
        condition, calculate fragment size, use local poisson to get
        pvalues and BH process to get qvalues, then combine 4-ways result
        to call differential sites.

        This script can use upto 4 cpus to speed up 4-ways calculation. (
        I am trying multiprocessing in python. )

        * MACS2/Constants.py, MACS2/IO/cBedGraph.pyx,
        MACS2/IO/cScoreTrack.pyx, MACS2/OptValidator.py,
        MACS2/PeakModel.py, MACS2/cPeakDetect.pyx

        All above files are modified for the new macs2diff script.

        * bin/macs2, bin/macs2diff, MACS2/OptValidator.py

        Now q-value 0.01 is the default cutoff. If -p is specified,
        p-value cutoff will be used instead.

2011-07-25  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.6 (tag:alpha)

        * bin/macsdiff

        A script to call differential regions. A naive way is introduced
        to find the regions where:

        1. signal from condition 1 is larger than input 1 and condition 2 --
        unique region in condition 1;
        2. signal from condition 2 is larger than input 2 and condition 1
        -- unique region in condition 2;
        3. signal from condition 1 is larger than input 1, signal from
        condition 2 is larger than input 2, however either signal from
        condition 1 or 2 is not larger than the other.

        Here 'larger' means the pvalue or qvalue from a Poisson test is
        under certain cutoff.

        (I will make another script to wrap up mulitple scripts for
        differential calling)

2011-07-07  Tao Liu  <vladimir.liu@gmail.com>
        MACS version 2.0.5 (tag:alpha)

        * bin/macs2, MACS2/cPeakDetect.py, MACS2/IO/cScoreTrack.pyx,
        MACS2/IO/cPeakIO.pyx

        Use hash to store peak information. Add back the feature to deal
        with data without control.

        Fix bug which incorrectly allows small peaks at the end of
        chromosomes.

        * bin/bdgpeakcall, bin/bdgcmp

        Fix bugs. bdgpeakcall can output encodePeak format.

2011-06-22  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.4 (tag:alpha)

        * cPeakDetect.py

        Fix a bug, correctly assign lambda_bg while --to-small is
        set. Thanks Junya Seo!

        Add rank and num of bp columns to pvalue-qvalue table.

        * cScoreTrack.py

        Fix bugs to correctly deal with peakless chromosomes. Thanks
        Vaibhav Jain!

        Use AFDR for independent tests instead.

        * encodePeak

        Now MACS can output peak coordinates together with pvalue, qvalue,
        summit positions in a single encodePeak format (designed for
        ENCODE project) file. This file can be loaded to UCSC
        browser. Definition of some specific columns are: 5th:
        int(-log10pvalue*10), 7th: fold-change, 8th: -log10pvalue, 9th:
        -log10qvalue, 10th: relative summit position to peak start.


2011-06-19  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.3 (tag:alpha)

        * Rich output with qvalue, fold enrichment, and pileup height

        Calculate q-values using a refined Benjamini–Hochberg–Yekutieli
        procedure:

        http://en.wikipedia.org/wiki/False_discovery_rate#Dependent_tests

        Now we have a similiar xls output file as before. The differences
        from previous file are:

        1. Summit now is absolute summit, instead of relative summit
           position;
        2. 'Pileup' is previous 'tag' column. It's the extended fragment
           pileup at the peak summit;
        3. We now use '-log10(pvalue)' instead of '-10log10(pvalue)', so
           5.00 means 1e-5, simple and less confusing.
        4. FDR column becomes '-log10(qvalue)' column.
        5. The pileup, -log10pvalue, fold_enrichment and -log10qvalue are
           the values at the peak summit.

        * Extra output files

        NAME_pqtable.txt contains pvalue and qvalue relationships.

        NAME_treat_pvalue.bdg and NAME_treat_qvalue.bdg store -log10pvalue
        and -log10qvalue scores in BedGraph format. Nearby regions with
        the same value are not merged.

        * Separation of FeatIO.py

        Its content has been divided into cPeakIO.pyx, cBedGraph.pyx, and
        cFixWidthTrack.pyx. A modified bedGraphTrackI class was
        implemented to store pileup, local lambda, pvalue, and qvalue
        alltogether in cScoreTrack.pyx.

        * Experimental option --half-ext

        Suggested by NPS algorithm, I added an experimental option
        --half-ext to let MACS only extends ChIP fragment around its
        middle point for only 1/2 d.

2011-06-12  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS version 2.0.2 (tag:alpha)

        * macs2

        Add an error check to see if there is no common chromosome names
        from treatment file and control file

        * cPeakDetect.pyx, cFeatIO.pyx, cPileup.pyx

        Reduce memory usage by removing deepcopy() calls.

        * Modify README documents and others.

2011-05-19  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS Version 2.0.1 (tag:alpha)

        * cPileup.pyx, cPeakDetect.pyx and peak calling process

        Jie suggested me a brilliant simple method to pileup fragments
        into bedGraph track. It works extremely faster than the previous
        function, i.e, faster than MACS1.3 or MACS1.4. So I can include
        large local lambda calculation in MACSv2 now. Now I generate three
        bedGraphs for d-size local bias, slocal-size and llocal-size local
        bias, and calculate the maximum local bias as local lambda
        bedGraph track.

        Minor: add_loc in bedGraphTrackI now can correctly merge the
        region with its preceding region if their value are the same.

        * macs2

        Add an option to shift control tags before extension. By default,
        control tags will be extended to both sides regardless of strand
        information.

2011-05-17  Tao Liu  <taoliu@jimmy.harvard.edu>
        MACS Version 2.0.0 (tag:alpha)

        * Use bedGraph type to store data internally and externally.

        We can have theoretically one-basepair resolution profiles. 10
        times smaller in filesize and even smaller after converting to
        bigWig for visualization.

        * Peak calling process modified. Better peak boundary detection.

        Extend ChIP tag to d, and pileup to have a ChIP bedGraph. Extend
        Control tag to d and 1,000bp, and pileup to two bedGraphs. (1000bp
        one will be averaged to d size) Then calculate the maximum value
        of these two tracks and a global background, to have a
        local-lambda bedGraph.

        Use -10log10poisson_pvalue as scores to generate a score track
        before peak calling.

        A general peak calling based on a score cutoff, min length of peak
        and max gap between nearby peaks.

        * Option changes.

        Wiggle file output is removed. Now we only support bedGraph
        output. The generation of bedGraph is highly recommended since it
        will not cost extra time. In other words, bedGraph generation is
        internally run even you don't want to save bedGraphs on disk, due
        to the peak calling algorithm in MACS v2.

        * cProb.pyx

        We now can calculate poisson pvalue in log space so that the score
        (-10*log10pvalue) will not have a upper limit of 3100 due to
        precision of float number.

        * Cython is adopted to speed up Python code.

2011-02-28  Tao Liu  <taoliu@jimmy.harvard.edu>
        Small fixes

        * Replaced with a newest WigTrackI class and fixed the wignorm script.

2011-02-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0rc2 (Valentine)

        * --single-wig option is renamed to --single-profile

        * BedGraph output with --bdg or -B option.

        The BedGraph output provides 1bp resolution fragment pileup
        profile. File size is smaller than wig file. This option can be
        combined with --single-profile option to produce a bedgraph file
        for the whole genome. This option can also make --space,
        --call-subpeaks invalid.

        * Fix the description of --shiftsize to correctly state that the
        value is 1/2 d (fragment size).

        * Fix a bug in the call to __filter_w_control_tags when control is
        not available.

        * Fix a bug on --to-small option. Now it works as expected.

        * Fix a bug while counting the tags in candidate peak region, an
        extra tag may be included. (Thanks to Jake Biesinger!)

        * Fix the bug for the peaks extended outside of chromosome
        start. If the minus strand tag goes outside of chromosome start
        after extension of d, it will be thrown out.

        * Post-process script for a combined wig file:

        The "wignorm" command can be called after a full run of MACS14 as
        a postprocess. wignorm can calculate the local background from the
        control wig file from MACS14, then use either foldchange,
        -10*log10(pvalue) from possion test, or difference after asinh
        transformation as the score to build a single wig track to
        represent the binding strength. This script will take a
        significant long time to process.

        * --wigextend has been obsoleted.

2010-09-21  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0rc1 (Starry Sky)

        * Duplicate reads option

        --keep-dup behavior is changed. Now user can specify how many
        reads he/she wants to keep at the same genomic location. 'auto' to
        let MACS decide the number based on binomial distribution, 'all'
        to let MACS keep all reads.

        * pvalue and FDR fixes (Thanks to Prof. Zhiping Weng)

        By default, MACS will now scale the smaller dataset to the bigger
        dataset. For instance, if IP has 10 million reads, and Input has 5
        million, MACS will double the lambda value calculated from Input
        reads while calling BOTH the positive peaks and negative
        peaks. This will address the issue caused by unbalanced numbers of
        reads from IP and Input. If --to-small is turned on, MACS will
        scale the larger dataset to the smaller one. So from now on, if d
        is fixed, then the peaks from a MACS call for A vs B should be
        identical to the negative peaks from a B vs A.

2010-09-01  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0beta (summer wishes)

        * New features

        ** Model building

        The default behavior in the model building step is slightly
        changed. When MACS can't find enough pairs to build model
        (implemented in alpha version) or the modeled fragment length is
        less than 2 times of tag length (implemented in beta version),
        MACS will use 2 times of --shiftsize value as fragment length in
        the later analysis. --off-auto can turn off this default behavior.

        ** Redundant tag filtering

        The IO module is rewritten. The redundant tag filtering process
        becomes simpler and works as promise. The maximum allowed number
        of tags at the exact same location is calculated from the
        sequencing depth and genome size using a binomial distribution,
        for both TREAMENT and CONTROL separately. ( previously only
        TREATMENT is considered ) The exact same location means the same
        coordination and the same strand. Then MACS will only keep at most
        this number of tags at the exact same location in the following
        analysis. An option --keep-dup can let MACS skip the filtering and
        keep all the tags. However this may bring in a lot of sequencing
        bias, so you may get many false positive peaks.

        ** Single wiggle mode

        First thing to mention, this is not the score track that I
        described before. By default, MACS generates wiggle files for
        fragment pileup for every chromosomes separately. When you use
        --single-wig option, MACS will generate a single wiggle file for
        all the chromosomes so you will get a wig.gz for TREATMENT and
        another wig.gz for CONTROL if available.

        ** Sniff -- automatic format detection

        Now, by default or "-f AUTO", MACS will decide the input file
        format automatically. Technically, it will try to read at most
        1000 records for the first 10 non-comment lines. If it succeeds,
        the format is decided. I recommend not to use AUTO and specify the
        right format for your input files, unless you combine different
        formats in a single MACS run.

        * Options changes

        --single-wig and --keep-dup are added. Check previous section in
        ChangeLog for detail.

        -f (--format) AUTO is now the default option.

        --slocal default: 1000
        --llocal default: 10000

        * Bug fixed

        Setup script will stop the installation if python version is not
        python2.6 or python2.7.

        Local lambda calculation has been changed back. MACS will check
        peak_region, slocal( default 1K) and llocal (default 10K) for the
        local bias. The previous 200bps default will cause MACS misses
        some peaks where the input bias is very sharp.

        sam2bed.py script is corrected.

        Relative pos in xls output is fixed.

        Parser for ELAND_export is fixed to pass some of the no match
        lines. And elandexport2bed.py is fixed too. ( however I can't
        guarantee that it works on any eland_export files. )

2010-06-04  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4.0alpha2 (be smarter)

        * Options changes

        --gsize now provides shortcuts for common genomes, including
        human, mouse, C. elegans and fruitfly.

        --llocal now will be 5000 bps if there is no input file, so that
        local lambda doesn't overkill enriched binding sites.

2010-06-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.4alpha (be smarter)
        
        * Options changes

        --tsize option is redesigned. MACS will use the first 10 lines of
        the input to decide the tag size. If user specifies --tsize, it
        will override the auto decided tsize.

        --lambdaset is replaced by --slocal and --llocal which mean the
        small local region and large local region. 

        --bw has no effect on the scan-window size now. It only affects the
        paired-peaks model process. 
        
        * Model building

        During the model building, MACS will pick out the enriched regions
        which are not too high and not too low to build the paired-peak
        model. Default the region is from fold 10 to fold 30. If MACS
        fails to build the model, by default it will use the nomodel
        settings, like shiftsize=100bps, to shift and extend each
        tags. This behavior can be turned off by '--off-auto'.

        * Output files

        An extra file including all the summit positions are saved in
        *_summits.bed file. An option '--call-subpeaks' will invoke
        PeakSplitter developed by Mali Salmon to split wide peaks into
        smaller subpeaks.
        
        * Sniff ( will in beta )

        Automatically recognize the input file format, so use can combine
        different format in one MACS run.

        Not implemented features/TODO:
        
        * Algorithms ( in near future? )

        MACS will try to refine the peak boundaries by calculating the
        scores for every point in the candidate peak regions. The score
        will be the -10*log(10,pvalue) on a local poisson distribution. A
        cutoff specified by users (--pvalue) will be applied to find the
        precise sub-peaks in the original candidate peak region. Peak
        boudaries and peak summits positions will be saved in separate BED
        files.

        * Single wiggle track ( in near future? )

        A single wiggle track will be generated to save the scores within
        candidate peak regions in the 10bps resolution. The wiggle file
        is in fixedStep format.


2009-10-16  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.7.1 (Oktoberfest, bug fixed #1)
        
        * bin/Constants.py

        Fixed typo. FCSTEP -> FESTEP

        * lib/PeakDetect.py

        The 'femax' attribute bug is fixed

2009-10-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.7 (Oktoberfest)
        
        * bin/macs, lib/PeakDetect.py, lib/IO/__init__.py, lib/OptValidator.py

        Enhancements by Peter Chines:

        1. gzip files are supported. 
        2. when --diag is on, user can set the increment and endpoint for
        fold enrichment analysis by setting --fe-step and --fe-max.

        Enhancements by Davide Cittaro:

        1. BAM and SAM formats are supported.
        2. small changes in the header lines of wiggle output.

        Enhancements by Me:
        1. I added --fe-min option;
        2. Bowtie ascii output with suffix ".map" is supported.
        
        Bug fixed:

        1. --nolambda bug is fixed. ( reported by Martin in JHU )
        2. --diag bug is fixed. ( reported by Bogdan Tanasa )
        3. Function to remove suffix '.fa' is fixed. ( reported by Jeff Johnston )
        4. Some "fold change" have been changed to "fold enrichment".

2009-06-10  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.6.1 (default parameter change)
        
        * bin/macs, lib/PeakDetect.py

        "--oldfdr" is removed. The 'oldfdr' behaviour becomes
        default. "--futurefdr" is added which can turn on the 'new' method
        introduced in 1.3.6. By default it's off.

        * lib/PeakDetect.py

        Fixed a bug. p-value is corrected a little bit.
        

2009-05-11  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.6 (Birthday cake)
        
        * bin/macs

        "track name" is added to the header of BED output file.

        Now the default peak detection method is to consider 5k and 10k
        nearby regions in treatment data and peak location, 1k, 5k, and
        10k regions in control data to calculate local bias. The old
        method can be called through '--old' option.

        Information about how many total/unique tags in treatment or
        control will be saved in final .xls output.

        * lib/IO/__init__.py

        ".fa" will be removed from input tag alignment so only the
        chromosome names are kept.

        WigTrackI class is added for Wiggle like data structure. (not used
        now)

        The parser for ELAND multi PET files has been fixed. Now the 5'
        tag position for a pair will be kept, whereas in the previous
        version, the middle points are kept.

        * lib/IO/BinKeeper.py

        BinKeeperI class is inspired by Jim Kent's library for UCSC genome
        browser, which can quickly access certain region for values in a
        large wiggle like data file. (not used now)

        * lib/OptValidator.py

        typo fixed.

        * lib/PeakDetect.py

        Now the default peak detection method is to consider 5k and 10k
        nearby regions in treatment data and peak location, 1k, 5k, and
        10k regions in control data to calculate local bias. The old
        method can be called through '--old' option.

        Two columns have beed added to BED output file. 4th column: peak
        name; 5th column: peak score using -10log(10,pvalue) as score.

        * setup.py

        Add support to build a Mac App through 'setup.py py2app', or a
        Windows executable through 'setup.py py2exe'. You need to install
        py2app or py2exe package in order to use these functions.

2009-02-12  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.5 (local lambda fixed, typo fixed, model figure improved)
        
        * PeakDetect.py

        Now, besides 1k, 5k, 10k, MACS will also consider peak size region
        in control data to calculate local lambda for each peak. Peak
        calling results will be slightly different with previous version,
        beware!

        * OptValidator.py

        Typo fixed, ELANDParser -> ELANDResultParser

        * OutputWriter.py

        Now, modeled d value will be shown on the model figure.

2009-01-06  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.4 (Happy New Year Version, bug fixed, ELAND multi/PET support)
        
        * macs, IO/__init__.py, PeakDetect.py

        Add support for ELAND multi format. Add support for Pair-End
        experiment, in this case, 5'end and 3'end ELAND multi format files
        are required for treatment or control data. See 00README file for
        detail.

        Add wigextend option.

        Add petdist option for Pair-End Tag experiment, which is the best
        distance between 5' and 3' tags.

        * PeakDetect.py

        Fixed a bug which cause the end positions of every peak region
        incorrectly added by 1 bp. ( Thanks Mali Salmon!)

        * OutputWriter.py
        
        Fix bugs while generating wiggle files. The start position of
        wiggle file is set to 1 instead of 0.

        Fix a bug that every 10M bps, signals in the first 'd' range are
        lower than actual. ( Thanks Mali Salmon!)


2008-12-03  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.3 (wiggle bugs fixed)
        
        * OutputWriter.py

        Fix bugs while generating wiggle files. 1. 'span=' is added to
        'variableStep' line; 2. previously, every 10M bps, the coordinates
        were wrongly shifted to the right for 'd' basepairs.

        * macs, PeakDetect.py

        Add an option to save wiggle files on different resolution.
        
2008-10-02  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.2 (tiny bugs fixed)

        * IO/__init__.py
        
        Fix 65536 -> 65535. ( Thank Joon) 
        
        * Prob.py
        
        Improved for binomial function with extra large number. Imported
        from Cistrome project.
        
        * PeakDetect.py

        If treatment channel misses reads in some chromosome included in
        control channel, or vice versa, MACS will not exit. (Thank Shaun
        Mahony)

        Instead, MACS will fake a tag at position -1 when calling
        treatment peaks vs control, but will ignore the chromosome while
        calling negative peaks.

2008-09-04  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3.1 (tiny bugs fixed version)

        * Prob.py
        
        Hyunjin Gene Shin contributed some codes to Prob.py. Now the
        binomial functions can tolerate large and small numbers.
        
        * IO/__init__.py

        Parsers now split lines in BED/ELAND file using any
        whitespaces. 'track' or 'browser' lines will be regarded as
        comment lines. A bug fixed when throwing StrandFormatError. The
        maximum redundant tag number at a single position can be no less
        than 65536.

        
2008-07-15  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.3 (naming clarification version)

        * Naming clarification changes according to our manuscript:

        'frag_len' is changed to 'd'.

        'fold_change' is changed to 'fold_enrichment'.
        
        Suggest '--bw' parameter to be determined by users from the real
        sonication size.

        Maximum FDR is 100% in the output file.

        And other clarifications in 00README file and the documents on the
        website.
        
        * IO/__init__.py
        If the redundant tag number at a single position is over 32767,
        just remember 32767, instead of raising an overflow exception.
        
        * setup.py
        fixed a typo.

        * PeakDetect.py
        Bug fixed for diagnosis report.
        

2008-07-10  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.2.2gamma

        * Serious bugs fix: 

        Poisson distribution CDF and inverse CDF functions are
        corrected. They can produce right results even for huge lambda
        now. So that the p-value and FDR values in the final excel sheet
        are corrected.

        IO package now can tolerate some rare cases; ELANDParser in IO
        package is fixed. (Thank Bogdan)

        * Improvement:

        Reverse paired peaks in model are rejected. So there will be no
        negative 'frag_len'. (Thank Bogdan)

        * Features added:
        
        Diagnosis function is completed. Which can output a table file for
        users to estimate their sequencing depth.


2008-06-30  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.2
        
        * Probe.py is added!  

        GSL is totally removed from MACS. Instead, I have implemented the
        CDF and inverse CDF for poisson and binomial distribution purely
        in python.

        * Constants.py is added!

        Organize constants used in MACS in the Constants.py file.

        * All other files are modified!

        Foldchange calculation is modified. Now the foldchange only be
        calculated at the peak summit position instead of the whole peak
        region. The values will be higher and more robust than before.

        Features added:

        1. MACS can save wiggle format files containing the tag number at
        every 10 bp along the genome. Tags are shifted according to our
        model before they are calculated.

        2. Model building and local lambda calculation can be skipped with
        certain options.

        3. A diagnosis report can be generated through '--diag'
        option. This report can help you get an assumption about the
        sequencing saturation. This funtion is only in beta stage.

        4. FDR calculation speed is highly improved.
        
2008-05-28  Tao Liu  <taoliu@jimmy.harvard.edu>
        Version 1.1
        
        * TabIO, PeakModel.py ...
        Bug fixed to let MACS tolerate some cases while there is no tag on
        either plus strand or minus strand.

        * setup.py
        Check the version of python. If the version is lower than 2.4,
        refuse to install with warning.