Expanding drug targets for 112 chronic diseases using a machine learning-assisted genetic priority score

ML-GPS: Machine Learning-Assisted Genetic Priority Score This Zenodo repository contains data and code associated with the publication: Chen R, Duffy Á, Petrazzini BO, Vy HM, Stein D, Mort M, Park JK, Schlessinger A, Itan Y, Cooper DN, Jordan DM, Rocheleau G, Do R. Expanding drug targets for 112 chronic diseases using a machine learning-assisted genetic priority score. Nat Commun. 2024 Oct 15;15(1):8891. doi: 10.1038/s41467-024-53333-y. Important notes You can interactively view the top 10% of ML-GPS predictions without download at https://rstudio-connect.hpc.mssm.edu/mlgps/. For running Jupyter notebooks, please follow the instructions in the README of the GitHub repository at https://github.com/robchiral/ML-GPS. Repository contents Files needed to train ML-GPS and ML-GPS DOE: Files needed for Jupyter notebooks.zip: Data files required for preprocessing and training. Jupyter notebooks.zip: Notebooks for cleaning data, training models, and generating predictions. Other files: Predictions for all gene-phecode pairs.zip: ML-GPS and ML-GPS DOE scores for all analyzed gene-phecode pairs. Summary statistics.zip: Genetic association summary statistics for all tested gene-phecode pairs. Updated performance metrics Model Open Targets AUPRC SIDER AUPRC ML-GPS (non-DOE) 0.074 0.080 ML-GPS DOE (activator predictions) 0.029 0.042 ML-GPS DOE (inhibitor predictions) 0.067 0.064 Zenodo versions Version 4: Updated notebooks and external data to use Open Targets 2024.9; summary statistics are unchanged Version 3: Corrected error where DOE for rare and ultrarare variants was incorrectly incorporated Version 2: Original release accompanying the publication