HTSeq to calculate strand specific coverage of pe reads

[moser]

I try to get strand specific coverage for paired-end RNA reads (using the first read in pair as the counting strand) from a sam-file (sorted by readname).
So far i found a solution using R, but for larger files (starting from 5 Gbyte), its just not feasible to do it with R as it takes ages.
By looking at HTSeq which can handle paired-end reads as bundles, i thought i could get strand specific coverage using the HTSeq.pair_SAM_alignments() method but didnt succeed so far...

import HTSeq
import pysam
import itertools

sam_reader = HTSeq.SAM_Reader( "test.sam")

cover = HTSeq.GenomicArray( "auto", stranded=True, typecode='i' )

for bundle in HTSeq.pair_SAM_alignments( sam_reader, bundle=True ):
        if len(bundle) != 1:
                continue
        left_almnt, right_almnt = bundle[0]  # extract pair
        if not left_almnt.aligned and right_almnt.aligned:
                continue

        cover[ left_almnt.iv ] += 1 

       #left_almnt equals left read from 5' end of pair NOT first read in pair

cover.write_bedgraph_file( "test-plus.wig", "+" )
cover.write_bedgraph_file( "test-minus.wig", "-" )

This gives me the strand specific coverage of the LEFT reads in pair.

I would like to:
- Find FIRST read in pair
- get its strand direction and change its mate's (LAST read in pair) strand direction to that of the FIRST read in pair
-count coverage of FIRST and LAST read in pair according to strand direction

Could anyone help me help with this task?
Maybe someone could show how to find FIRST read in pair with HTSeq?

Thank you,
Michel

[dpryan]

Why bother scripting it when htseq-count already does that?

[moser]

I thought htseq-count is for counting reads in features.
I would like to have per base coverage not counts per feature.
Or you mean by feeding a per-base-gff file as feature and do the counting to get coverage?

[dpryan]

Ah, I misunderstood what you wanted. I'm not sure that HTSeq provides all of the interface that you want. I'm sure you can do that strictly with pysam, though, at least unless your reads are coordinate sorted.