MapLight TDC

This repository contains the source code for MapLight's Therapeutics Data Commons (TDC) ADMET Benchmark Group submission.

Installation

This codebase describes MapLight's two submissions to the TDC leaderboards:

1. MapLight model (submission.ipynb): CatBoost gradient boosted decision trees with ECFP, Avalon, and ErG fingerprints, as well as 200 physicochemical descriptors. Runnable on Colab.

2. MapLight + GNN model (submission_gnn.ipynb): the same as the MapLight model with graph isomorphism network (GIN) supervised masking fingerprints from molfeat. WARNING: Not runnable on Colab becuase of this issue.

Both notebooks will install all dependencies in a new Python environment with JupyterLab:

# Create an environment for this project
mamba create -n maplight python=3.10 -y && mamba activate maplight

mamba install jupyterlab