Announcement

Collapse

Welcome to the New Seqanswers!

Welcome to the new Seqanswers! We'd love your feedback, please post any you have to this topic: New Seqanswers Feedback.
See more
See less

inconsistancy between cufflinks and cuffdiffs output?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by masterpiece View Post
    "So tell me, which one gives the best result and explain me why (answer not more than 50 words and use only layman term)?"
    Haha, this might have been the best thing I've seen on SeqAnswers.

    There's also DEXSeq, which does splicing, so at least you can have some consistency in your method. Whatever you do, don't stick with 1.4, as even the authors will tell you, that version was not good.

    One thing you mentioned about lab work, I'm assuming you mean qPCR. How do you handle it with fold changes less than 2? I get many genes that DESeq says are differentially expressed, and the fold change might be as low as 1.25, and no one in my lab seems to think I can validate that with qPCR.

    Comment


    • #17
      Originally posted by billstevens View Post

      One thing you mentioned about lab work, I'm assuming you mean qPCR. How do you handle it with fold changes less than 2? I get many genes that DESeq says are differentially expressed, and the fold change might be as low as 1.25, and no one in my lab seems to think I can validate that with qPCR.
      May I know do you have replicates in your analysis. Without replicates, the statistics shown will not be strong enough to support your results.

      Normally in my case, apart from using p-value or FDR to get a list of differently expressed genes, I'll also include in log2foldchange set at +/-2. Maybe you should try include log2foldchange and see how many differently expressed genes left.

      Comment


      • #18
        Yes, I definitely have replicates.

        If I only kept log2foldchange set at 2, I would lose the majority of my genes, and DESeq becomes much less powerful.

        I guess the fundamental question I'm asking here is, if you can't do it verify via qPCR, should you call the gene differentially expressed. And if so, then the "power" of all of these software programs is kinda useless.

        Comment


        • #19
          billstevens

          Unfortunately I don't think I can help you on this. My task is generating results from sequence reads, run the statistics analysis. Later i pass them to other person in our lab for validation test through qPCR. I my self don't have any experience running the test. And I'm not sure if genes with fold change as low as 1.25 can be validated through qPCR.

          As in our case we set p-val = 0.05 and log2foldchange =+/-2 for the genes to be differently express, so we don't face the problem that you have mentioned.

          Comment


          • #20
            Hi guys,
            After reading through the discussion above I am realizing that I should also try DESeq. I have two multi-replicate samples and ran cuffdiff on them. Cuffdiff produces a read count files in output that has three types of read counts (actual, internally adjusted andexternlly adjusted) for each transcript in each replicate. Which one should I use for DESeq?

            Comment


            • #21
              vinay052003,

              As far I know, you can't use cuffdiff output to be used in DEseq run. You have to start from bam files generated by tophat and run through Htseq-count and DEseq

              tophat -> Htseq-count -> DEseq

              For more details on how to run the DEseq you can refer threads below and others in seqanswer

              http://seqanswers.com/forums/showthr...ht=htseq-count

              http://seqanswers.com/forums/showthr...ht=htseq-count
              Last edited by masterpiece; 10-09-2012, 07:48 PM.

              Comment


              • #22
                Cufflinks/cuffdiff, etc. outputs garbage - don't trust it - don't use it. I have spent a considerable amount of time checking outputs against the raw alignment files and the number and magnitude of errors is quite astounding. Use HTseq-count and DEseq instead.

                Comment

                Working...
                X