help calculating unique mapping rate for paired end

[emolinari]

Hi Everyone!

I am writing down some stats for my RNAseq data.
Briefly, I have used Tophat2+samtools+HTSeq count for mapping and gene counting (PE, GRCh37).

When i check the samtools flagstat everything looks terrific...great mapping rates as well as pairing between reads.

I was trying to get my unique mapping rates stats:
I've been using the accepted_hits.bam file and run

samtools view accepted_hits.bam | cut -f1 | sort | uniq | wc -l


My naive question is:
the number I get, is it considering the pair or the number of reads???

Anyone has a better solution for counting unique mapping rate for PE?

Thanks!

Manu

[atcghelix]

If you're talking about unique with regards to mapping positions on both ends, how about marking duplicates using Picard's MarkDuplicates tool, then using samtools flagstat? You can subtract the duplicate reads from the mapped reads to get the number of non-duplicate mapped reads.

[dpryan]

I would strongly discourage you from marking duplicates in RNAseq data, as the alleged duplicates will primarily be false positives.

Anyway, tophat2 will set the MAPQ to 0,1,2, or 50 (or something like that) depending on the number of best hits a read/pair has. If an alignment has a MAPQ of 50, then it maps uniquely (for some definitions of "uniquely mapping"). So just "samtools view -c -q 40 foo.bam" for the number of alignments. This counts both reads in a pair separately, so divide by 2 for an approximate number of read pairs ("approximate" in the case of singletons).