LigandNet is a Machine Learning (ML) toolbox that combines different ligand-based models into an open source platform that can predict if a ligand may have an activity to a specific protein. Finding a ligand that will bind to a human protein and have a significant signaling effect through the cell can be an expensive task. Nowadays, ML models are being employed throughout many scientific fields and are being successfully applied to drug discovery research. In this project, we have applied advanced ML approaches such as Random Forest (RF), Support Vector Machine (SVM), Linear Regression (LR), Extra Tree Classifier (ETC) and Deep Learning (DL) to classify the ligands as active/binder or inactive/nonbinder. We obtained the known active ligands for each of 1704 proteins from Pharos (pharos.nih.gov) database. For each of the known active ligands, decoys were generated using DecoyFinder (http://urvnutrigenomica-ctns.github.io/DecoyFinder/) and Zinc database (http://zinc.docking.org/). ML models were developed for each of the protein-ligand sets by using known actives and generated decoys. ECFP6 fingerprints and Topological Pharmacophore Atomic Triplet Fingerprint (TPATF) from MayaChemTools (http://www.mayachemtools.org/) were employed as feature generators in developing the models. Models were validated using highest positive predictive value (PPV), sensitivity, and area under the curve of the receiver operating characteristic plots (ROC-AUC) to determine which model works best with each dataset of the proteins, determining the accuracy and precision of each ML approach. The developed models are available on GitHub (https://github.com/sirimullalab/LigandNet).<br/>

LigandNet: A machine-learning-based toolkit for predicting ligand activity to proteins

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.