[dpryan]

looks like most of my alignment has ~60% properly paired reads. I wonder if there is way to filter the bam bile and separate the paired reads/singleton... to feed the dexseq...

Was that meant to be a challenge ?

#!/usr/bin/env python
import csv
import sys

f = csv.reader(sys.stdin, dialect="excel-tab")
of = csv.writer(sys.stdout, dialect="excel-tab")
previous = None
for line in f :
    if((int(line[1]) & 0x100) != 0) :
        #Don't output secondary alignments
        continue 
    if((int(line[1]) & 0x2) == 0) :
        #Skip anything not properly paired
        continue
    if(previous != None and line[0] == previous[0]) :
        of.writerow(previous)  
        of.writerow(line)
        previous = None
    else :
        previous = line

Usage would be something like:

samtools view namesorted.bam | filter.py | dexseq_count.py ....

That will filter out anything that's not properly paired, has is a secondary alignment, or is missing its mate. You can use that as a basis if you want to filter in a different way.