Consensus gene model

[ChrisAU]

Hi all,

I have three different transcript annotations in gff format, two of them were generated by Cufflinks. Now I would like to build a consensus gene model out of the three. Do you have any suggestions, which programs are suited best for this task?

Thanks
Chris

[RockChalkJayhawk]

Cufflinks should also report gene expression for you. Otherwise I would use mergeBed from BEDtools to build a consensus.

[ChrisAU]

I have two gene models from Cufflinks and a third one. They are quite similar for the most part, however, in some regions the exon-intron structure differs. The objective is, to calculate the most probable gene model out of them. mergeBed just merges all structures, there's no validation of the predicted models involved.

[arvid]

You could try letting RSEM assign the original reads to the sequences of your candidate transcripts. It will fit the reads by a EM method and thus assign the reads to the transcripts in a way that fits the data... Afterwards you could drop the models with low read support.

If the differences are minimal, it will be difficult to verify the models computationally - I would go back to the lab and clone the transcripts.