Predictive performance on selected GPCR target receptors
(A) Predictive performance of the proposed LISA-CPI on 6 selected top-20 GPCR targets. Predicted pKi and ground-truth pKi of each compound for each GPCR target are contour plotted with point density. Pearson’s correlation coefficient R and p values are labeled.
(B) Predictive performance of the proposed LISA-CPI on 8 selected pain-associated GPCR targets. Predicted pKi and ground-truth pKi of each compound for each GPCR target are contour plotted with point density. Pearson’s correlation coefficient R and p values are labeled.
(C) Receiver operating characteristic (ROC) curves showcasing the predictive performance of the proposed LISA-CPI and three other models (ImageMol, CHEM-BERT, and MolCLR) on 2 selected kinase targets and the entire kinase dataset. Solid lines and shades represent the mean and one standard deviation of ROC curves obtained from 10-fold cross-validation, respectively.