pseudoalign call to analyze_reads not gathering correct info for kallisto about arcashla

[IoanFilip2]

Thank you for your interest in our tool, and for this informative post. The observation is correct: what is required is the fragment length/std and not the read length, which is what we are supplying for single-end reads. That is an error, and fixing it could also improve performance due to omitting the running of analyze_reads(...). However, it appears that fragment length information is only used in kallisto quantification to account for sequence-specific biases as a result of nonrandom priming of fragments (see kallisto paper). In fact, we have our own quantification step in arcasHLA that does not directly use kallisto quant. Additionally, in kallisto quantification step, the mean/std are only used to generate weights from a Gaussian distribution with these specified parameters. We expect the impact of using mean/std of reads as opposed to mean/std of fragments to be minimal, if any at all, on our results.