I do a lot of manipulation to BAM/SAM file formats and came across some questions regarding how to filter these results efficiently. Below, I have a python script that I use to filter based on the reference_name section of the BAM file. Note: I work in Metagenomics so many times I am aligning reads to a genome scaffold, which contains many different contigs. One application of this tool would be to filter out a BAM file to only contain contigs that were placed in a metagenome-assembled genome bin.

import sys
import argparse
import pysam


def readBinID(binid, keep_bin=False):
    contigList = [0] * 400000
    with open(binid) as b:
        line = b.readline().strip()
        while line:
            if keep_bin:
                bin_name = line.split('\t')[0]
                if bin_name in keep_bin:
                    contig = int(line.split('\t')[1].split('_')[1])
                    contigList[contig] = 1
            else:
                contig = int(line.split('\t')[1].split('_')[1])
                contigList[contig] = 1
            line = b.readline().strip()
    return contigList


def filterBam(inBam, outBam, contigs, keep_bin=False):
    contigList = readBinID(contigs, keep_bin)
    bam = pysam.AlignmentFile(inBam)
    obam = pysam.AlignmentFile(outBam, 'wb', template=bam)

    for b in bam.fetch(until_eof=True):
        try:
            index = int(b.reference_name.split('_')[1])
        except AttributeError:
            pass
        if contigList[index] == 1:
            obam.write(b)

    bam.close()
    obam.close()
    return 0


if __name__ == '__main__':
    from datetime import datetime
    start = datetime.now()
    parser = argparse.ArgumentParser(description="Parser")
    parser.add_argument("-i", "--Input",
                        help="Input .BAM file to filter.",
                        default=sys.stdin, required=False)
    parser.add_argument("-o", "--Output",
                        help="Output .BAM file to write to.",
                        required=True)
    parser.add_argument("-c", "--Contigs", help="File containing reference \
    identifiers and the associated bin (each on its own line & tab-separated)",
                        required=True)
    parser.add_argument("-k", "--Keep",
                        help="Keep only contigs in the specified bins.",
                        nargs="*", required=False, default=False)
    argument = parser.parse_args()
    filterBam(argument.Input, argument.Output,
              argument.Contigs, keep_bin=argument.Keep)
    msg = f"{datetime.now() - start}\n"
    with open('log.txt', 'a') as log:
        log.write(msg)

The idea here is that the first function, readBinID, just takes in a simple tab-delimited file containing all of the contigs of interest. This file has 2 fields: bin the contig belongs in, and the contig name. This could instead be easily manipulated to be a file containing a list of all the read ids, for example, and then the filterBam function would just need to change from 'b.reference_name' to 'b.query_name' I believe. There's functionality in this for selecting all contigs in the file or specifying based on being in a certain bin.

What's most import here is initializing a list full of 0s and finding a way to code each contig so you know where each is in the list. With the programs I work with, my contigs are easily labeled from 1 - last contig, so pretty direct.

For a 2Gb bam file containing reads mapped to ~300,000 different references, it takes around 1-2 minutes at most to filter with 1 CPU (memory requirements are super low).

If anyone has any ideas on how to speed this up I'd greatly appreciate it.

Hope this may help someone.

DGD