search

TobiTekath / DTUrtle

Perform differential transcript usage (DTU) analysis of bulk or single-cell RNA-seq data. See documentation at:
https://tobitekath.github.io/DTUrtle
GNU General Public License v3.0
17 stars 3 forks source link

Use of covariates in DTUrtle #1

Open antgomo opened 3 years ago

antgomo commented 3 years ago

Hi,

i was wondering if i can use covariates in the analysis (i.e Sex, age or technical ones)

Kind Regards

TobiTekath commented 3 years ago

Hi @antgomo ,

thank you for your interest in DTUrtle.

Currently, only classical case-control designs are supported, but the support for more complex designs is definitely on the list for future releases.

In the meantime, you could try to provide an 'adjusted' count matrix to DTUrtle, for example with limma's removeBatchEffect or with the specialised package ComBat-seq.

We would really appreciate, if you could share your experience using DTUrtle with us. Thanks a lot.

Best, Tobias

yoavhadas commented 11 months ago

@TobiTekath, what is your opinion on using Harmony or Seurats integration anchors approach to correct batch effects on the transcript matrix? How would you recommend using the corrected matrix in DTUrle?

TobiTekath commented 11 months ago

@yoavhadas Very interesting question.

First of all, I do not have tested any integration approach or the effect on the DTU calls - so I can only rely on my gut feeling here.

This said, I would have slight concern directly applying Harmony or Seurats anchor integration on the transcript expression matrices, as there methods do not take the gene - transcript relationship into account. So each transcript's expression values could slightly be "normalized" independently, thus just by chance changing the ratios and creating a DTU signal (same applies for the mentionend bulk batchremoval methods above).

Again, taking my limited tests regarding batch effects into account, I am wondering if the need for batch normalization is as urgent for DTU analysis, as we are comparing proportional changes in the end (which are invariant of linear expression differences). Also I wonder how one would "correct" proportions which could be explained by a covariate, without losing true signal. More than happy to hear your thoughts :)