Chemi-Net: A Molecular Graph Convolutional Network for Accurate Drug Property Prediction

TLDR

ADME studies are critical for drug discovery and are traditionally performed using domain‑specific feature descriptors or fingerprints. The authors developed Chemi‑Net, a data‑driven, domain‑knowledge‑free deep learning model, to predict ADME properties. They evaluated Chemi‑Net against Amgen’s Cubist program by conducting a large‑scale ADME prediction study on‑site at Amgen. Chemi‑Net achieved higher R² values than Cubist across 13 datasets, with a median 26.7 % improvement, promising accelerated drug discovery.

Abstract

Absorption, distribution, metabolism, and excretion (ADME) studies are critical for drug discovery. Conventionally, these tasks, together with other chemical property predictions, rely on domain-specific feature descriptors, or fingerprints. Following the recent success of neural networks, we developed Chemi-Net, a completely data-driven, domain knowledge-free, deep learning method for ADME property prediction. To compare the relative performance of Chemi-Net with Cubist, one of the popular machine learning programs used by Amgen, a large-scale ADME property prediction study was performed on-site at Amgen. For all 13 data sets, Chemi-Net resulted in higher R2 values compared with the Cubist benchmark. The median R2 increase rate over Cubist was 26.7%. We expect that the significantly increased accuracy of ADME prediction seen with Chemi-Net over Cubist will greatly accelerate drug discovery.