Skip to content

A python extension for quick access to bigWig and bigBed files

License

Notifications You must be signed in to change notification settings

tturowski/pyBigWig

 
 

Repository files navigation

PyPI version Travis-CI status bioconda-badge DOI

pyBigWig

A python extension, written in C, for quick access to bigBed files and access to and creation of bigWig files. This extension uses libBigWig for local and remote file access.

Table of Contents

Installation

You can install this extension directly from github with:

pip install pyBigWig

or with conda

conda install pybigwig -c conda-forge -c bioconda

Requirements

The follow non-python requirements must be installed:

  • libcurl (and the curl-config config)
  • zlib

The headers and libraries for these are required.

Usage

Basic usage is as follows:

Load the extension

>>> import pyBigWig

Open a bigWig or bigBed file

This will work if your working directory is the pyBigWig source code directory.

>>> bw = pyBigWig.open("test/test.bw")

Note that if the file doesn't exist you'll see an error message and None will be returned. Be default, all files are opened for reading and not writing. You can alter this by passing a mode containing w:

>>> bw = pyBigWig.open("test/output.bw", "w")

Note that a file opened for writing can't be queried for its intervals or statistics, it can only be written to. If you open a file for writing then you will next need to add a header (see the section on this below).

Local and remote bigBed read access is also supported:

>>> bb = pyBigWig.open("https://www.encodeproject.org/files/ENCFF001JBR/@@download/ENCFF001JBR.bigBed")

While you can specify a mode for bigBed files, it is ignored. The object returned by pyBigWig.open() is the same regardless of whether you're opening a bigWig or bigBed file.

Determining the file type

Since bigWig and bigBed files can both be opened, it may be necessary to determine whether a given bigWigFile object points to a bigWig or bigBed file. To that end, one can use the isBigWig() and isBigBed() functions:

>>> bw = pyBigWig.open("test/test.bw")
>>> bw.isBigWig()
True
>>> bw.isBigBed()
False

Access the list of chromosomes and their lengths

bigWigFile objects contain a dictionary holding the chromosome lengths, which can be accessed with the chroms() accessor.

>>> bw.chroms()
dict_proxy({'1': 195471971L, '10': 130694993L})

You can also directly query a particular chromosome.

>>> bw.chroms("1")
195471971L

The lengths are stored a the "long" integer type, which is why there's an L suffix. If you specify a non-existant chromosome then nothing is output.

>>> bw.chroms("c")
>>> 

Print the header

It's sometimes useful to print a bigWig's header. This is presented here as a python dictionary containing: the version (typically 4), the number of zoom levels (nLevels), the number of bases described (nBasesCovered), the minimum value (minVal), the maximum value (maxVal), the sum of all values (sumData), and the sum of all squared values (sumSquared). The last two of these are needed for determining the mean and standard deviation.

>>> bw.header()
{'maxVal': 2L, 'sumData': 272L, 'minVal': 0L, 'version': 4L, 'sumSquared': 500L, 'nLevels': 1L, 'nBasesCovered': 154L}

Note that this is also possible for bigBed files and the same dictionary keys will be present. Entries such as maxVal, sumData, minVal, and sumSquared are then largely not meaningful.

Compute summary information on a range

bigWig files are used to store values associated with positions and ranges of them. Typically we want to quickly access the average value over a range, which is very simple:

>>> bw.stats("1", 0, 3)
[0.2000000054637591]

Suppose instead of the mean value, we instead wanted the maximum value:

>>> bw.stats("1", 0, 3, type="max")
[0.30000001192092896]

Other options are "min" (the minimum value), "coverage" (the fraction of bases covered), and "std" (the standard deviation of the values).

It's often the case that we would instead like to compute values of some number of evenly spaced bins in a given interval, which is also simple:

>>> bw.stats("1",99, 200, type="max", nBins=2)
[1.399999976158142, 1.5]

nBins defaults to 1, just as type defaults to mean.

If the start and end positions are omitted then the entire chromosome is used:

>>> bw.stats("1")
[1.3351851569281683]

A note on statistics and zoom levels

A note to the lay reader: This section is rather technical and included only for the sake of completeness. The summary is that if your needs require exact mean/max/etc. summary values for an interval or intervals and that a small trade-off in speed is acceptable, that you should use the exact=True option in the stats() function.

By default, there are some unintuitive aspects to computing statistics on ranges in a bigWig file. The bigWig format was originally created in the context of genome browsers. There, computing exact summary statistics for a given interval is less important than quickly being able to compute an approximate statistic (after all, browsers need to be able to quickly display a number of contiguous intervals and support scrolling/zooming). Because of this, bigWig files contain not only interval-value associations, but also sum of values/sum of squared values/minimum value/maximum value/number of bases covered for equally sized bins of various sizes. These different sizes are referred to as "zoom levels". The smallest zoom level has bins that are 16 times the mean interval size in the file and each subsequent zoom level has bins 4 times larger than the previous. This methodology is used in Kent's tools and, therefore, likely used in almost every currently existing bigWig file.

When a bigWig file is queried for a summary statistic, the size of the interval is used to determine whether to use a zoom level and, if so, which one. The optimal zoom level is that which has the largest bins no more than half the width of the desired interval. If no such zoom level exists, the original intervals are instead used for the calculation.

For the sake of consistency with other tools, pyBigWig adopts this same methodology. However, since this is (A) unintuitive and (B) undesirable in some applications, pyBigWig enables computation of exact summary statistics regardless of the interval size (i.e., it allows ignoring the zoom levels). This was originally proposed here and an example is below:

>>> import pyBigWig
>>> from numpy import mean
>>> bw = pyBigWig.open("http://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeMapability/wgEncodeCrgMapabilityAlign75mer.bigWig")
>>> bw.stats('chr1', 89294, 91629)
[0.20120902053804418]
>>> mean(bw.values('chr1', 89294, 91629))
0.22213841940688142
>>> bw.stats('chr1', 89294, 91629, exact=True)
[0.22213841940688142]

Retrieve values for individual bases in a range

While the stats() method can be used to retrieve the original values for each base (e.g., by setting nBins to the number of bases), it's preferable to instead use the values() accessor.

>>> bw.values("1", 0, 3)
[0.10000000149011612, 0.20000000298023224, 0.30000001192092896]

The list produced will always contain one value for every base in the range specified. If a particular base has no associated value in the bigWig file then the returned value will be nan.

>>> bw.values("1", 0, 4)
[0.10000000149011612, 0.20000000298023224, 0.30000001192092896, nan]

Retrieve all intervals in a range

Sometimes it's convenient to retrieve all entries overlapping some range. This can be done with the intervals() function:

>>> bw.intervals("1", 0, 3)
((0, 1, 0.10000000149011612), (1, 2, 0.20000000298023224), (2, 3, 0.30000001192092896))

What's returned is a list of tuples containing: the start position, end end position, and the value. Thus, the example above has values of 0.1, 0.2, and 0.3 at positions 0, 1, and 2, respectively.

If the start and end position are omitted then all intervals on the chromosome specified are returned:

>>> bw.intervals("1")
((0, 1, 0.10000000149011612), (1, 2, 0.20000000298023224), (2, 3, 0.30000001192092896), (100, 150, 1.399999976158142), (150, 151, 1.5))

Retrieving bigBed entries

As opposed to bigWig files, bigBed files hold entries, which are intervals with an associated string. You can access these entries using the entries() function:

>>> bb = pyBigWig.open("https://www.encodeproject.org/files/ENCFF001JBR/@@download/ENCFF001JBR.bigBed")
>>> bb.entries('chr1', 10000000, 10020000)
[(10009333, 10009640, '61035\t130\t-\t0.026\t0.42\t404'), (10014007, 10014289, '61047\t136\t-\t0.029\t0.42\t404'), (10014373, 10024307, '61048\t630\t-\t5.420\t0.00\t2672399')]

The output is a list of entry tuples. The tuple elements are the start and end position of each entry, followed by its associated string. The string is returned exactly as it's held in the bigBed file, so parsing it is left to you. To determine what the various fields are in these string, consult the SQL string:

>>> bb.SQL()
table RnaElements
"BED6 + 3 scores for RNA Elements data"
    (
    string chrom;      "Reference sequence chromosome or scaffold"
    uint   chromStart; "Start position in chromosome"
    uint   chromEnd;   "End position in chromosome"
    string name;       "Name of item"
    uint   score;      "Normalized score from 0-1000"
    char[1] strand;    "+ or - or . for unknown"
    float level;       "Expression level such as RPKM or FPKM. Set to -1 for no data."
    float signif;      "Statistical significance such as IDR. Set to -1 for no data."
    uint score2;       "Additional measurement/count e.g. number of reads. Set to 0 for no data."
    )

Note that the first three entries in the SQL string are not part of the string.

If you only need to know where entries are and not their associated values, you can save memory by additionally specifying withString=False in entries():

>>> bb.entries('chr1', 10000000, 10020000, withString=False)
[(10009333, 10009640), (10014007, 10014289), (10014373, 10024307)]

Add a header to a bigWig file

If you've opened a file for writing then you'll need to give it a header before you can add any entries. The header contains all of the chromosomes, in order, and their sizes. If your genome has two chromosomes, chr1 and chr2, of lengths 1 and 1.5 million bases, then the following would add an appropriate header:

>>> bw.addHeader([("chr1", 1000000), ("chr2", 1500000)])

bigWig headers are case-sensitive, so chr1 and Chr1 are different. Likewise, 1 and chr1 are not the same, so you can't mix Ensembl and UCSC chromosome names. After adding a header, you can then add entries.

By default, up to 10 "zoom levels" are constructed for bigWig files. You can change this default number with the maxZooms optional argument. A common use of this is to create a bigWig file that simply holds intervals and no zoom levels:

>>> bw.addHeader([("chr1", 1000000), ("chr2", 1500000)], maxZooms=0)

If you set maxTooms=0, please note that IGV and many other tools WILL NOT WORK as they assume that at least one zoom level will be present. You are advised to use the default unless you do not expect the bigWig files to be used by other packages.

Adding entries to a bigWig file

Assuming you've opened a file for writing and added a header, you can then add entries. Note that the entries must be added in order, as bigWig files always contain ordered intervals. There are three formats that bigWig files can use internally to store entries. The most commonly observed format is identical to a bedGraph file:

chr1	0	100	0.0
chr1	100	120	1.0
chr1	125	126	200.0

These entries would be added as follows:

>>> bw.addEntries(["chr1", "chr1", "chr1"], [0, 100, 125], ends=[5, 120, 126], values=[0.0, 1.0, 200.0])

Each entry occupies 12 bytes before compression.

The second format uses a fixed span, but a variable step size between entries. These can be represented in a wiggle file as:

variableStep chrom=chr1 span=20
500	-2.0
600	150.0
635	25.0

The above entries describe (1-based) positions 501-520, 601-620 and 636-655. These would be added as follows:

>>> bw.addEntries("chr1", [500, 600, 635], values=[-2.0, 150.0, 25.0], span=20)

Each entry of this type occupies 8 bytes before compression.

The final format uses a fixed step and span for each entry, corresponding to the fixedStep wiggle format:

fixedStep chrom=chr1 step=30 span=20
-5.0
-20.0
25.0

The above entries describe (1-based) bases 901-920, 931-950 and 961-980 and would be added as follows:

>>> bw.addEntries("chr1", 900, values=[-5.0, -20.0, 25.0], span=20, step=30)

Each entry of this type occupies 4 bytes.

Note that pyBigWig will try to prevent you from adding entries in an incorrect order. This, however, requires additional over-head. Should that not be acceptable, you can simply specify validate=False when adding entries:

>>> bw.addEntries(["chr1", "chr1", "chr1"], [100, 0, 125], ends=[120, 5, 126], values=[0.0, 1.0, 200.0], validate=False)

You're obviously then responsible for ensuring that you do not add entries out of order. The resulting files would otherwise largley not be usable.

Close a bigWig or bigBed file

A file can be closed with a simple bw.close(), as is commonly done with other file types. For files opened for writing, closing a file writes any buffered entries to disk, constructs and writes the file index, and constructs zoom levels. Consequently, this can take a bit of time.

Numpy

As of version 0.3.0, pyBigWig supports input of coordinates using numpy integers and vectors in some functions if numpy was installed prior to installing pyBigWig. To determine if pyBigWig was installed with numpy support by checking the numpy accessor:

>>> import pyBigWig
>>> pyBigWig.numpy
1

If pyBigWig.numpy is 1, then pyBigWig was compiled with numpy support. This means that addEntries() can accept numpy coordinates:

>>> import pyBigWig
>>> import numpy
>>> bw = pyBigWig.open("/tmp/delete.bw", "w")
>>> bw.addHeader([("1", 1000)], maxZooms=0)
>>> chroms = np.array(["1"] * 10)
>>> starts = np.array([0, 10, 20, 30, 40, 50, 60, 70, 80, 90], dtype=np.int64)
>>> ends = np.array([5, 15, 25, 35, 45, 55, 65, 75, 85, 95], dtype=np.int64)
>>> values0 = np.array(np.random.random_sample(10), dtype=np.float64)
>>> bw.addEntries(chroms, starts, ends=ends, values=values0)
>>> bw.close()

Additionally, values() can directly output a numpy vector:

>>> bw = bw.open("/tmp/delete.bw")
>>> bw.values('1', 0, 10, numpy=True)
[ 0.74336642  0.74336642  0.74336642  0.74336642  0.74336642         nan
     nan         nan         nan         nan]
>>> type(bw.values('1', 0, 10, numpy=True))
<type 'numpy.ndarray'>

Remote file access

If you do not have curl installed, pyBigWig will be installed without the ability to access remote files. You can determine if you will be able to access remote files with pyBigWig.remote. If that returns 1, then you can access remote files. If it returns 0 then you can't.

Empty files

As of version 0.3.5, pyBigWig is able to read and write bigWig files lacking entries. Please note that such files are generally not compatible with other programs, since there's no definition of how a bigWig file with no entries should look. For such a file, the intervals() accessor will return None, the stats() function will return a list of None of the desired length, and values() will return [] (an empty list). This should generally allow programs utilizing pyBigWig to continue without issue.

For those wishing to mimic the functionality of pyBigWig/libBigWig in this regard, please note that it looks at the number of bases covered (as reported in the file header) to check for "empty" files.

A note on coordinates

Wiggle, bigWig, and bigBed files use 0-based half-open coordinates, which are also used by this extension. So to access the value for the first base on chr1, one would specify the starting position as 0 and the end position as 1. Similarly, bases 100 to 115 would have a start of 99 and an end of 115. This is simply for the sake of consistency with the underlying bigWig file and may change in the future.

Galaxy

pyBigWig is also available as a package in Galaxy. You can find it in the toolshed and the IUC is currently hosting the XML definition of this on github.

About

A python extension for quick access to bigWig and bigBed files

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C 93.2%
  • Python 6.8%