Figure 5
Distribution of theoretical \(R_{0\ test}^{2}\) (A) and RMSE test (B) values. The mean of the \(R_{0\ test}^{2}\) distribution, 0.68, highlights that the uncertainty in public bioactivity data does not permit models displaying \(R_{0\ test}^{2}\) values close to 1. Similar results were obtained for \(q^{2}_{\textit {test}}\). From these data we conclude that the minimum RMSEtest value that a model can achieve without exhibiting overfitting is close to the experimental uncertainty.