Cuffdiff results with frag-bias-correct vary considerably when re-runing
Dear all,
I facing a very strange issue when running the tuxedo workflow with wheat data.
The problem is that I fail to reproduce my own results when I run cuffdiff with the bias correction mode. What I mean is that if I re-run cuffdiff using the same parameters (-b among them) and input files I may or may not get the same results. I sometimes get 63 differentially expressed (DE) genes whereas other times over 3000 genes are reported as significant.
I have realised that the difference between these two groups of results is in the number of genes that are not tested for differential expression, that is, the number of genes marked as “NOTEST”. Out of the 116,000 genes in the merged GTF file, sometimes 112,000 are not tested and other times that number decreases to 85,000, which is when I get a higher number of DE genes. I am completely confused with this behaviour and the only thing that I have clear is that the values in bias_params.info change when the results differ whilst those in var_model.info remain the same. It seems to me that what sometimes is deemed as good enough (OK) to test by cuffdiff is other times discarded (NOTEST). I have repeated the analysis several times, starting two times from the cufflinks step and repeating cuffdiff more than 10 times and I still cannot spot any reason for this pseudo-random behaviour.
I would appreciate any help, so let me know if you require any further information.
Thanks for all and best regards,
Asier
I facing a very strange issue when running the tuxedo workflow with wheat data.
The problem is that I fail to reproduce my own results when I run cuffdiff with the bias correction mode. What I mean is that if I re-run cuffdiff using the same parameters (-b among them) and input files I may or may not get the same results. I sometimes get 63 differentially expressed (DE) genes whereas other times over 3000 genes are reported as significant.
I have realised that the difference between these two groups of results is in the number of genes that are not tested for differential expression, that is, the number of genes marked as “NOTEST”. Out of the 116,000 genes in the merged GTF file, sometimes 112,000 are not tested and other times that number decreases to 85,000, which is when I get a higher number of DE genes. I am completely confused with this behaviour and the only thing that I have clear is that the values in bias_params.info change when the results differ whilst those in var_model.info remain the same. It seems to me that what sometimes is deemed as good enough (OK) to test by cuffdiff is other times discarded (NOTEST). I have repeated the analysis several times, starting two times from the cufflinks step and repeating cuffdiff more than 10 times and I still cannot spot any reason for this pseudo-random behaviour.
I would appreciate any help, so let me know if you require any further information.
Thanks for all and best regards,
Asier