RNA seq result interpretation

Question

[reena_gd]

Hello, I am doing RNAseq analysis for the first time. I have two samples, control and treatment of a plant variety collected at 14 days interval. I obtained few data sets of differentially expressed genes which had similar gene ID and were also the same transcripts. They were same except that, they differed in FPKM values and had different regulation, like one is up-regulated (12-fold) and other is down-regulated (13-fold). I assume that minor variation errors could be possible but such fold variation along with up-regulation and down-regulation cannot be overlooked. I also don’t think that they could be different fragments of the same transcript as they show different regulation. Can anyone suggest the reason for such data? Or is this mere an error.

[Abhishek]

Did you find a way out of this?

Answer 1

[Kevin Blighe]

Large log fold-changes are often observed in RNA-seq data that has undergone normalisation to FPKM expression levels, even as high as +90, but this is more due to the inadequacies of this normalisation strategy than anything else. For one, this normalisation is not performed across samples and is therefore not adequately adjusting for different library sizes.

If you can obtain raw counts, my advice is to get those, and then work from those using a 'better' normalisation strategy.

[reena_gd]

Thanks Kevin, I tried another strategy, but results are not much varying than previous. I suppose removing such ambiguous data would be better.

[Kevin Blighe]

Which was the other strategy? Have you checked for sample outliers via something like a PCA bi-plot?

[reena_gd]

I don't think these could be outliers because several other genes have similar up and down regulation values. I have

                                  GENE_MODEL_ID       RefSeq_ID                 control read count          treated read count
TCONS_00047959        XLOC_028640             XM_003535153.3          135.743                       0.00891439                                                                               
TCONS_00047960        XLOC_028640             XM_003535153.3         0.00996383                         70.1898

I expect TCONS ID differed because it is generated for each different transcript in each experiment

[Kevin Blighe]

The large fold-changes are obviously related to one have a very high value, while the other a very low value. It may make biological sense for one isoform to be only activated in the treatment condition, while the other [isoform] is deactivated, and vice-versa. Further research would be needed.