Fpkm differences between versions of Cufflinks

[zorph]

I have run the same paired-end SOLiD sample on both version of Cufflinks (0.9.3 and 1.0.3).

When looking at the transcripts.gtf file from both versions, I observe that the FPKMs from the same transcript are different.

1)Does anyone know what changed between the two versions to cause this difference?

When I tried to run the latest version of Cuffcompare (1.0.3) on my paired-end SOLiD sample, no FPKM values were reported--they were all zero.

2) does anyone know why this is the case and what can be done to correct this?


Thank You!

[arielpaulson]

Just noticed this old post, but since no-one had replied, figured why not:

Pre-1.2.0 Cufflinks did not calculate FPKMs in a deterministic manner -- perhaps the EM was randomly initialized -- you could run the same version of Cufflinks on the same dataset 10 times and get 10 different FPKMs.

Granted, the variance would be small, but occasionally it would not. I've calculated that about 1% of the time the FPKM confidence intervals between runs would not overlap (for the same gene).

Anyway, this problem has been fixed as of 1.2.0.

Ariel