Cufflinks merging nearby genes

[xinchen]

Hi everyone,

I'm running Cufflinks (v 0.9.3) on one of my datasets (human, 50 bp PE, around ~600-700 million reads), and tried to assemble transcripts without prior annotation.

For the most part, the program works very well. However, when genes adjacent to one another (intergenic distance on the order of a small/medium-sized intron) are the same strand, Cufflinks merge multiple genes into one transcript. I've attached a picture of this example, where DVL2 & PHF23 are both part of the same Cufftranscript.

Would anyone have any recommendations to fix this or reduce the chances of it happening? Since the intergenic distance is short, tweaking the "max-intron-length" flag in Cufflinks probably won't help here.

Thanks!

[lpachter]

This problem is most likely being caused by polymerase run off on the 3' end of the PHF23 gene.

It is possible, in principle, to identify two distinct genes from such data if the coverage is substantially different. However at this time the Cufflinks assembler does not explicitly take such information into account (it does in phasing alternative transcripts).

A new forthcoming feature in Cufflinks to be released in the next version is a "reference guided assembly" where you can inform the assembler based on existing annotation. In this case, the resulting assembly would contain the two known genes thus preventing your problem. The nice thing is that the reference guided assembler will still allow for the discovery of new transcripts. Coming soon.

[xinchen]

Thanks! It definitely looks like polymerase run-off, so it'll be great to use the new Cufflinks feature for using reference assemblies to separate these merged genes.