4 ms·
carbocation: Can you say a word or two about the current view of short reads for re-sequencing? I worked for a time on software for a particular lab to do ali
by dasht 16y ago
carbocation:
Can you say a word or two about the current view of short reads for re-sequencing? I worked for a time on software for a particular lab to do alignment of short(-ish) reads like 13-bp+gap-of-100-200+7bp+gap-of-5-8+7bp+gap-of-100-200+13-bp. The idea was that you give me millions or more of those reads, and a reference like the human genome, and I give you back all of the alignments including partial matches (a few bp off, here and there).
Interesting programming problem. This aren't the actual bp counts and gap sizes I was working with - they were in that range. I got concerned that the reads we were working with were too ambiguous -- similar to the problem you describe with de novo. I was having trouble doing the combinatorics to prove myself wrong and started asking: hey, where's the mathematical analysis that shows reads of this size can possibly work? Nobody could produce it and this led to some internal strife.
That was a few years back. What's current thinking on what kind of reads plausibly work for re-sequencing? (If you happen to know. And where can I read something about the math that proves it.)
- carbocation 16y agoYou probably know more about the intricate details than I do! I think perhaps one of the more important recent approaches for potentially gapped or slightly mismatched reads is the use of the Burrows-Wheeler transform. http://en.wikipedia.org/wiki/Burrows–Wheeler_transform http://en.wikipedia.org/wiki/Burrows–Wheeler_transform Heng Li wrote BWA based on this, and it seems to handle some of the issues you were wondering about.
- aheilbut 16y agoIt's more an empirical question rather than a theoretical one, and short-read sequencing wouldn't be as popular as it is if it didn't work out most of the time. I don't have all the exact numbers handy, but 76bp will uniquely map about 80-90% of potential reads, just with single ends. A high proportion of reads can be mapped uniquely even down to 25 or 30bp. There are some 'mappability' tracks available in the UCSC browser if you're curious about a particular region. http://hgdownload.cse.ucsc.edu/goldenPath/hg18/encodeDCC/wgEncodeMapability/ http://hgdownload.cse.ucsc.edu/goldenPath/hg18/encodeDCC/wgE... Of course there are repetitive and low complexity regions that can't be sequenced at all with these read lengths, or which haven't yet been properly sequenced or assembled at all even with much longer reads. I'm not aware of a paper that has fully explicated how read-length and the structure of the reads affects mappability (especially with pairs or reads with multiple gaps), though it's certainly something that all the people building the instruments (and the mapping algorithms) have thought about a quite a lot.