EM algorithm in Kallisto

Question

[Mozart]

Dear staff, This is the second time I try to post. The original one was a bit wishy-washy so I'll try my best to be clear.

I know Kallisto harnesses the EM algorithm during the pseudoalignment step. Am I correct saying that, using the same pseudoalignment parameters for the same experiment, the number of rounds during the EM algorithm step are sample-specific and therefore number of rounds are different for different samples? I assume this is the case as the EM algorithm refers to transcript contained in each sample, that are biologically different for every sample. It is always good to double check, tho.

Thanks

Answer 1

[dsull]

Hi,

Yes, your biological samples (or technical bootstrap samples) will get you different number of rounds.

I wouldn't call it "sample-specific" -- but if you feed an EM algorithm different data with the same initialization parameters, it'll go down a different path (and hence, terminate differently).

The EM algorithm simply runs until a termination criteria is met (when the log likelihood changes less than some value).

[Mozart]

This was exactly what I hoped to hear to confirm I got this right.

Thanks a lot for that.

Have a good day