Samtools Pileup Of Millions Of Reads From A Single Amplicon

Question

[Sebastian]

Hi all,

We would like to pileup millions of reads from a single amplicon for ultra-sensitive mutation detection.

Considering that SAMtools pileup is limited to several thousand reads at a given position I am wondering if you could suggest us any alternative approach or workaround.

Any feedback is highly appreciated!

[Christian]

If you sequence so deep, how do you make sure that not 99% of your reads are pcr duplicates?

[Sebastian]

In fact they are PCR duplicates, as the reads derive from amplicons. But that's not a problem, we want to detect single reads out of more than one million having a specific (known) mutation.

[Pierre Lindenbaum]

"But that's not a problem, ee want to detect single reads out of more than one million having a specific (known) mutation." So you just want to have a look at the cigar string of each read , isn't it ?

[Sebastian]

That is one additional approach we already thought about. But with millions of reads we also have to think about sequencing errors, both at the mutation and flanking sites. So our mutation may be expressed by different CIGAR strings. Further, e.g. any 4-base mutation at the same position will result in the same CIGAR string. As a third point, there may be the need to detect unknown mutations in a known hotspot in the future as well, that is why we need a flexible approach....

[Pierre Lindenbaum]

shuffle & downsample your bam ? or are you just searching for the reads having a SNP ?

[Sebastian]

yes, we want to detect those reads having a mutation

[Pierre Lindenbaum]

Cross-posted on Samtools mailing list http://sourceforge.net/mailarchive/forum.php?thread_name=20854588711E4A489A3AD70C9BA5548A01AE472348A7%40XCH11.scidom.de&forum_name=samtools-help

[Pierre Lindenbaum]

and crossposted on http://seqanswers.com/forums/showthread.php?t=41050

Answer 1

[Christian]

Could you throw away all reads that match the reference at your hotspot and pileup the rest? This should massively decrease your read count.

[Sebastian]

Unfortunately, this is only a possibility when I expect a really low frequency. Usually, I don't know the frequency of my sample and it might be upt o 50%. Furthermore, I would expect lots of wild-type reads that contain sequencing errors and thus also do not map the reference....

[Christian]

The first issue you could address with downsampling. The second one mitigated by checking base qualities.