theislab / diffxpy

Differential expression analysis for single-cell RNA-seq data.
https://diffxpy.rtfd.io
BSD 3-Clause "New" or "Revised" License
193 stars 23 forks source link

Documentation on reported log2fc when using continous predictor variables #199

Open Hrovatin opened 3 years ago

Hrovatin commented 3 years ago

I have run a DE test where I have multiple continuous predictor variables (one of interest and multiple covariates). Sometimes I got really low reported log2fc (that would be filtered out if using usual DE filtering thresholds) and other times they were "normal". I figured out that the difference is because I once scaled my variable of interest to [0,1] while the other time it was approx 10x bigger; other covariates had the same range in both cases. This of course affected the fitted coefficient and thus reported log2fc, which seems to be the tilt of the line, right? Maybe this could be added in documentation (if not already and I just overlooked it) or the "log2fc" column in summary could be renamed in cases when it is in fact not log2fc.