The code for running each model is divided into individual sub-folders. Two types of model execution can be done: 1) run a pretrained model with specified hyperparameters; 2) run a model from scratch with specified hyperparameters. The former execution can be done by running the run_pretrained.sh script and the latter can be done by running run_model_with_hyp.sh script. Hyperparameter tuning has been performed on the validation set and the best set of hyperparameters for each validation strategy (leave-ccls-out/LCO, leave-drugs-out/LDO, leave-pairs-out /LPO) and each pathway collection (KEGG, PID, Reactome) are provided in sub-folders named best_hyp. All pathway-based models (PathDNN, ConsDeepSignaling, HiDRA, PathDSP) are re-implementations of the original models, with a very small component of code being adaptations (direct usage) of the original code provided by the authors of these pathway-based models. References for such adaptations are included in the comments of the code.