Why large negative values with logT = False?

[aarmey]

Why large negative values with logT = False?

[mcreixell]

I'm attaching an example following an instance where I'm getting large negative values before logT (idx 13 in your screenshot). The large negative values come from the mean center results. I am a bit confused about your comments on the last PR. In the second comment it seemed like the code looked good to you as it is but in the first comment you mention I should be mean-centering on the log-transformed data instead of the linear values and divide by the mean instead of subtract it? Also I'm calculating the fold-change by dividing by the control instead of subtracting it as you suggest. Sorry for the misunderstanding.

@aarmey:
by mean-centering, I meant doing so on the log-transformed data (in effect, this is dividing by the mean). Then going back to fold-change is actually subtracting the control so it's 0.0 (1.0 outside of log). So, all the data should just remain in log-space throughout.

[aarmey]

You should mean-center the log-transformed values. What I meant to convey is this operation is identical to dividing linear-scale data by the geometric mean (https://en.wikipedia.org/wiki/Geometric_mean). I'm not suggesting to do the latter, though. Sorry that wasn't clear.

[mcreixell]

Great, thanks.