BedTool object to pandas dataframe

[radaniba]

Hello

I am using pybedtool quite a lot and each time I find myself writing a small function to convert a BedTool object (let's say result of a coverage) to a pandas dataframe in order to use it for other purposes or to inject it into other functions

It would be great to have this as a prebuilt utility , I guess a lot of people would like to have it by default

Cheers

Rad

[daler]

Sure -- what does your current function look like?

[radaniba]

Well it depends on the result actually, let's say we have a result from genom_coverage per base :

so it is gonna be something like

def coverage_to_df(bam_file):
    coverage_result = bam_file.genome_coverage(d=True)
    #we initialize the dataframe that will contain the coverage result
    # This will be easier to use pd df for plotting and subsets selections
    df = pd.DataFrame(columns=['chrom', 'position', 'coverage'])
    row_id = 0
    for pos_cov in coverage_result:
        chrom, position, coverage = pos_cov.split('\t')
        df.loc[row_id] = [chrom, position, coverage]
        row_id = row_id + 1

But I imagine it is depending on the context and on the BedTool object that varies in term of number of columns

[radaniba]

I was thinking about something like

coverage_result = bam_file.genome_coverage(d=True).to_data_frame()

and the user is free to update the columns of his dataframe with the labels he wants or

coverage_result = bam_file.genome_coverage(d=True).to_data_frame(['chrom','position','coverage'])

with columns names as arguments, that way it will be practical to do the whole thing in one line and that will fit all BedTool objects

[daler]

For anything but a BAM file, you could just call pandas.read_table on the underlying filename (fn attribute):

import pybedtools
import pandas
x = pybedtools.example_bedtool('a.bed')
df = pandas.read_table(x.fn, names=['chrom', 'start', 'stop', 'name', 'score', 'strand'])

What's the speed like on incrementally building a dataframe from BAM coverage like in your example? I suspect it would be faster to just read the file in -- pandas' parsers are pretty fast.

Making this built-in would be trivial:

def to_dataframe(self, *args, **kwargs):
    """
    create a pandas.DataFrame, passing args and kwargs to pandas.read_table
    """
    # Complain if BAM or if not a file
    if self._isbam:
        raise ValueError("BAM not supported for converting to DataFrame")
    if not isinstance(self.fn, basestring):
        raise ValueError("use .saveas() to make sure self.fn is a file")

    # Otherwise we're good:
    return pandas.read_table(self.fn, *args, **kwargs)

Would this work for you? I suppose a lookup table mapping filetype/field count to default names values would help cut down on typing.

[radaniba]

Thanks @daler, yes that function is enough and it does what it is needed to do.

[radaniba]

awesome ! Thanks @daler