For missense variants on the TSC2 protein, the validation script prohibits negative predicted values. I understand that, as explained in the Wiki, "0 = unstable, 1 = wildtype, and >1 denotes predicted stability greater than wildtype". However, similar DMS experiments, such as the related challenges "PTEN and TPMT in CAGI5", display a distribution with some values below zero. Thus, a predictor targeting the same assay that in TSC2 is also expected to yield some negative values. A question thus naturally arises: when evaluating, each option seems to have a drawback, so are you going to: (1) clip negative experimental values to zero (thus losing some information in the process), (2) rescale the distribution around 1 to avoid negative values (thus leading to a different distribution from the ones in related challenges), (3) leave the experimental values as they are (thus comparing X with forced non-negative values to Y with negative values allowed), or (4) another option? A related question is: does it matter to replicate the experimental distribution, or will only variant ranking matter? Thank you and all the best, Matsvei Tsishyn