Hi all,

I'm starting to use the tximport package to pull in salmon output for statistical analysis and I just wanted to verify that it is fine to use the non-integer counts with edgeR and DESeq2. Years ago, we needed to round the decimal counts from e.g., cuffdiff, but the tximport vignette implied the non-integer values were fine because there was no mention of rounding. I searched the support site and found these: C: Can I feed TCGA normalized count data to EdgeR for differential gene expression and non-integer counts for edgeR, the first of which has Steve Lianoglou asking for a separate post to highlight the qualified yes answer in regards to non-integer counts for edgeR now. That's mainly what this post is doing, and also linking it to the tximport package as more people should now be using it to pull in counts and length offsets from kallisto/sailfish/salmon/RSEM.

BTW - thanks for the great tximport package and the F1000 article on how transcript-level estimates improve gene-level inferences: https://f1000research.com/articles/4-1521/v2!

Jenny

hi Jenny,

Yes the examples in the tximport vignette show the intended downstream usage.

edgeR has support for non-integer counts, and the tximport-to-DESeqDataSet constructor function just rounds the estimated non-integer counts to integers. I'm not worried about any loss of precision for inference of log fold change in this rounding, because fractions of counts are tiny compared to the sampling and biological variation on counts in RNA-seq.