Recommended settings for STARlong

[Katarina Truvé]

Dear Alex,

I have been running reads 2*126. (Quality looks real good) When using default settings (normal STAR) only 63% of reads were uniquely mapped. I tried STARlong default settings which helped a bit and raised the mapping to 74%. Most unmapped went into “unmapped:too short”. I continued and ran STARlong again changing from default settings to --outFilterScoreMinOverLread 0.3 --outFilterMatchNminOverLread 0.3. This now raised uniquely mapped to 95%. If I tried doing the same with "normal STAR" the mapping did not improve much but unmapped moved to reads “mapped to multiple loci”. So I’m happy with the 95% unique mapping!, my problem is that I do not know exactly what this change of default values does to the data. Is this change ok, or does it introduce error to the mapping? I also tried to find any advise about how to run STARlong, but couldn’t find anything in the manual. Are there any written instructions anywhere? Do you recommend any other changes in settings then the ones I  tried here to improve mapping?

Best,

Katarina

P.S Attached some data from the "best" final-log.

[Alexander Dobin]

Hi Katarina,

I do not recommend using the STARlong for short reads <200-300b. STARlong is less sensitive to the multimapping short reads, and this probably explains the increase in the "uniquely" mapping reads that you observed. In other words, if normal STAR reports that reads are multi-mappers, and STARlong marks them as unique mappers, you should believe normal STAR.

If you post (or send me) the Log.final.out files of the normal STAR runs with both default and reduced --outFilterScoreMinOverLread 0.3 --outFilterMatchNminOverLread 0.3, I could make some suggestions no how to investigate why the reads do not map, or map as multimappers.

Cheers

Alex