Hi,
I have noticed that the newer versions of Cufflinks are giving much more significant differentially expressed genes. I am updating my pipeline, so I am confused whether to use the new version or not. Has anyone validated any results from new version or have any idea about the false positive rate? Please let me know your thoughts and suggestions.
Difference in number of significant genes detected by Cufflinks :
Genes Isoforms Splicing
cufflinks_2.0.2 14 11 98
cufflinks_2.1.1 97 59 26
cufflinks_2.2.1 79 51 14
Thanks,
Sid
•
written
by
siddharth.sethi5
I would generally prefer DESeq among the popular RNA-Seq tools. I think working directly with the RPKM values is another good option.
You can see some benchmarks here.
•
written
by
Charles Warden
I use DESeq and and EdgeR as well. But I am also interested in the Isoforms and splicing differential expression, thats where I use cufflinks. Do you know about other popular tools for isoform and splicing. I am already using MATS and Bitseq.
MATS is my favorite. MISO is another popular option.
I would prefer the splicing event predictions in MATS or MISO to the whole-transcript predictions in tools like cufflinks, RSEM, etc (I would recommend sticking to gene-level quantification for gene expression analysis). At least for me, I didn't think visual inspection of the alignment looked suitable every time where I have seen a potentially interesting result where one transcript was supposed to behave very different than the other transcripts for the same gene. I think it is hard because there are a lot of factors in defining a given transcript and coverage (at least for the samples that I have seen) is often very uneven.
•
written
by
Charles Warden
Sailfish and RNA-Skim for transcript level RNA-seq quantification. Both are reference mapping free (no tophat) so they are very fast!
•
written
by
Ming Tommy Tang
Thanks very much guys for the help and suggestions.
I had the impression that different versions of cufflinks do give different results. You may try other methods like DESeq or EdgeR.
but for DESeq2 (better) also gives different results from DESeq. I do not think there is a "correct" way to analyze the data. As long as you have a hypothesis to test, and the results make sense given that no mistakes in analyzing them, it's OK.
Log in to answer this question.
