pFP/: script to compute atom-pair fingerprints per datapoint.
dPLEC/: script to compute close contact fingerprints between ligands and proteins (irrelevant for freesolv).
dFEAT/: script to compute molecular properties per datapoint.
datasets/: folder containing all datasets to train on. Also contains the python script that compiles them.
SVM/: contains the script to run the training protocol. Might have to remove tensorflow imports.
General workflow: compile all molecules into datasets/input/, then run the python scripts for all three feature generation types. Once these have finished running, run compile_datasets.py in datasets/ to combine all feature sets into different combinations. Once the datasets have been compiled, the builder script in SVM/ can be run to start training; the other script plots convergence (this needs to be rewritten as it assumes splits per congeneric series).