Tuning Sailfish Performance vs Salmon

[Billy Lau]

Salmon has a pretty nice feature where you can change -c and -k to tune the mapping %, which is useful for noisy reads. Are there similar options for Sailfish? I've tried using smaller k for building the index but anything below 20 (say, 15) just makes the program run like a snail.

Also, how should I cite Salmon as it looks like there isn't a preprint out yet? I'd really like to include results from it in my manuscript.

[Rob]

Hi Billy,

  The underlying methodologies of salmon and sailfish are significantly different.  The mapping % reported by salmon is a true mapping percentage based on the underlying fragments (i.e. reads, or read pairs in the case of paired-end data).  The mapping % reported by sailfish is a (highly-conservative) estimate of the mapping rate that is based on extrapolating the fraction of mapping k-mers to fragments.  This is almost always a vast under-estimate of the "effective" mapping rate (this is a calculation that we've been meaning to change for a while and we'll probably get around to it very soon --- as soon as the salmon pre-print is out).  This means that the amount of information being used by sailfish is likely substantially more than is indicated by the reported mapping %.

  However, that being said, we've developed salmon because we believe the underlying approach and methodology improves upon that of sailfish (e.g. as you mention, the -c  and -k parameters, enabled by our novel approach to lightweight-alignment,  provide a level of versatility not possible in sailfish).  We plan to have a pre-print for salmon out very soon.  However, for the time being, you can cite the Sailfish paper and mention the tool along the lines of "quantification was performed with Salmon (the successor to Sailfish [cite SF paper])".  This is what we've done so far in pre-prints and manuscripts of ours that make use of salmon.

Best,

Rob