Strange Extremely High FPKMs at Random

[tonup69]

I have been running the TOPHAT pipeline for analysis and have had a problem with some EXTREMELY high FPKM values for a few of the samples. Has anyone seen this before? What do these extremely high FPKM values mean? They appear to be artifactual and only appear on one biological replicate typically (although not always the same sample).

[kopi-o]

If you have used Cufflinks, you will often find such FPKM values for very short transcripts, like miRNAs. Check if that is the case.

[tonup69]

If you have used Cufflinks, you will often find such FPKM values for very short transcripts, like miRNAs. Check if that is the case.

How would you check that? Do you know how to select for genes with Log10FPKM>6 as a gene set? That would allow me to look at these things.

LTR

[baptist]

tonup69:

I think in the output directory of cuffdiff, there are files with suffix fpkm_tracking. Maybe you can get what you want from these files?

[gwilymh]

I have had a similar problem. These massively elevated FPKM values seem to only be reported for the transcripts newly discovered by Cufflinks/TopHat (i.e. transcripts that were not previously annotated).

In one analysis, for example, the known genes had an average FPKM value of 214.5, and a maximum FPKM value of 147,473. The newly discovered transcripts, however, returned a mean FPKM of 139,234.6 and a maximum of 74,769,200. 2.5% of the new transcripts had vales greater than the maximum FPKM detected for the annotated genes. The results for the new transcripts clearly contain artifacts.

Does anyone know how or why this occurs? How can it be prevented? How can such artifacts be screened out of downstream analyses?