I would like to download RNAseq gene expression data from GTEx from the liver only. What is the best way to do this?

I have tried a few things. I downloaded several files from the GTEx website (https://www.gtexportal.org/home/datasets, under "RNA-Seq Data"), but the problems I run into are:

  • Not sure what is the best file: "Gene read counts" or "Gene TPMs"? What is the difference?
  • I would like 1 value per sample per gene. There are many duplicate genes in these 2 files. How can I collapse them to 1 value per gene?
  • How can I download the data for liver tissue only, instead of ALL data for ALL tissues?

Thank you for suggestions.

  • 1
    $\begingroup$ What do you eventually want to do with the data? That helps deciding what to use. $\endgroup$
    – user3051
    Commented Apr 11, 2020 at 14:46
  • 1
    $\begingroup$ What do you mean by duplicates? I assume you mean multiple entries with the same gene name. Gene names are not unique identifiers. Also, whether you want raw counts or TPMs depends on what you want to do with the data. If you want to calculate differential expression between 2 samples, you will need counts. Honestly, comparing expression values of different genes is apples to oranges. Plus, you need replicates, ideally from the same batch. RNA-Seq is very noisy data. Finally, if you want a specific tissue, you have to find samples of that tissue. $\endgroup$
    – hepcat72
    Commented Apr 12, 2020 at 15:45
  • $\begingroup$ @ATpoint and hepcat72 Thank you. I have found that subsetting the data worked to get data from one tissue only. The gene names start with ENSGxxxx. I found out that they are different transcripts, sometimes of the same gene, so I suppose there are no real duplicates. $\endgroup$ Commented Apr 15, 2020 at 1:05

1 Answer 1

  • Gene read counts are the number of reads or fragments that are mapped to each gene. So it is an integer and the values are not directly comparable between samples due to differences in sequencing depth. TPM stands transcripts per million. It is the normalized gene expression level. Basically, you first normalize read counts by gene length and then normalize the library size to 1e6 in all libraries. As for which one should you use, it depends on what analysis you want to do and what tool you are going to use.
  • I have not noticed that there are duplicated genes in those files, can you show an example?
  • I do not think you can download expression from one tissue, but you can always subset. The whole dataset is not very large.
  • $\begingroup$ Thank you very much. Subsetting worked for me. $\endgroup$ Commented Apr 15, 2020 at 1:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.