Open madison-feshuk opened 2 weeks ago
Instead, create method resp.censormed.neg25 to remove all responses from concentrations which the median falls below -25.
New method to censor responses below -25% seems to be working great. See reprocessed Zearalenone example:
Will share ne output with PI for review.
Since we’ve switched to fitting everything bidirectionally, gain/loss responses in this proliferation/cytotox assay with an asymmetrical +100% gain/-300% loss are missed and we end up only fitting the loss. I don’t think we want to NOT fit bidirectionally and lose that cytotox response. “Overwriting” the hitcall based on intended direction also wouldn’t solve this.
The -300% is essentially an artificial response calculated relative to baseline, so we could try adding a correction method so that no values fall below -100%. The extreme values may be contributing to model fitting in the negative direction, so bounding should improve this.
Some examples with less extreme negative responses appear to accurately capture proliferation followed by cytotoxicity with biphasic poly2 and gnls models: