I give you guys a lot of credit for your transparency and for tackling solubility. Considering the many variables involved with determining the solubility of a molecule - version, form, assay protocol, etc. - your results, from someone who see the glass half-full, are somewhat encouraging. It seems very clear that expanding your training set will significantly improve predictions, but in the meantime, providing scientists with "confidence" metrics will provide the valuable context they need to adequately use the predictors.
I give you guys a lot of credit for your transparency and for tackling solubility. Considering the many variables involved with determining the solubility of a molecule - version, form, assay protocol, etc. - your results, from someone who see the glass half-full, are somewhat encouraging. It seems very clear that expanding your training set will significantly improve predictions, but in the meantime, providing scientists with "confidence" metrics will provide the valuable context they need to adequately use the predictors.