Hey,
we have done some experiments with our standard dUTP strand specific protocol which is NEBNext Ultra. Unfortunately, we don't have enough material to use this protocol for the next experiment which is why we want to use the SMARTer kit from Clontech. This can be problematic because we are using the same cell lines from the previous experiment and want to make sure that we don't introduce a heavy bias with another library preparation. I know there will be bias but I was wondering what's the best way to check the differences?
What we have done so far is, we took the two samples and prepped each of them twice. Once with NEXNext Ultra and once with SMARTer. These prepped samples were sequenced (up to 60 mio fragments paired-end). At then end we have 4 sequenced samples: A-normalPrep, A-SMARTer, B-normalPrep, B-SMARTer. Now I thought of comparing the samples by counting the fragments of each prep, loading the counts table into DESeq and using the normalised counts to produce a Spearman correlation (A-normalPrep vs A-SMARTer, the same for B). Another option would be creating FPKM counts but I don't like that because it's not meant for library-library comparison. Or I was wondering if Kallisto (http://pachterlab.github.io/kallisto/about.html) would be way to. What do you think? Any recommendations and advice is welcome.
Thanks
Mathias
I wonder what your problem actually is. Yes there will definitely be consistent differences between the two prep methods but I wonder why that is a problem for you. Is it for example the case that you have already sequenced a batch of samples with method A and in the future it will be method B and you want to use all your samples in, lets say a differential expression analysis. Then just use the prep methode as a covariate. Or do you want to quantify the correlation between methode a and b? Yes you can do what you suggested. You can also use DEseq normalized gene expression estimates and calculate pearson distance. How good do the correlations need to b? Say pearson correlation between normalized (Indeed dont use fpkm) is > 0.95 are you then satisfied?