search

COMBINE-lab / RapMap

Rapid sensitive and accurate read mapping via quasi-mapping
GNU General Public License v3.0
89 stars 23 forks source link

Add support for Salmon library types #2

Closed rob-p closed 5 years ago

rob-p commented 9 years ago

It would be nice to allow RapMap to mark alignments as concordant or discordant --- right now it has no idea of these concepts.

rob-p commented 9 years ago

Since we already have (from the QuasiAlignment itself) the manner in which each read maps, this is a low priority issue.

mdshw5 commented 8 years ago

Is there any plan to add QuasiAlignment output as an option for Salmon? I understand that this would be off by default, but would be useful as an option...

rob-p commented 8 years ago

This could be done fairly easily I think. There are two places I see this being useful. One is just to have the set of all quasi-mappings found by Salmon so you know what it was working with. The other related feature (I think @Blahah would be interested in this) would be to output disambiguated quasi-mappings after quantification has happened. This requires a second quasi-mapping pass through the reads (since remembering all mappings would waste too much space), but it would give you a nice set of alignments — one target per read — where the probability of assigning a multimapping read to a target is proportional to that target's inferred abundance. Salmon can already do with with alignments from a bam file, and it's useful in certain cases.

blahah commented 8 years ago

+10^999999...

rob-p commented 8 years ago

@Blahah — would the current RapMap output format (target / position / strand, encoded in SAM format) be useful to you at all yet, or would you need the full-blown alignment before this becomes useful?

blahah commented 8 years ago

We need some way to know the number of mismatches between the read and the target. Can that be calculated from the information already available without doing an alignment?

rob-p commented 8 years ago

@Blahah — not really. For example, we know the number and length of exactly matching intervals (maximum mappable prefixes) between the read and reference, but not what occurs in between. This means it would be possible to miss some matches. However, as we discussed before, actually providing the alignment should not be very difficult; it just hasn't been implemented yet. However, seeing the pain the aligner causes in Transrate, I'd like to make it an action item.