MEGA-CL: A Molecular Foundation Model for Generalizable ADMET Prediction through Graph External Attention and Contrastive Learning
For drug discovery researchers, MEGA-CL provides a generalizable and accurate ADMET prediction framework that demonstrates practical utility in early-stage candidate optimization.
MEGA-CL is a graph neural network foundation model for ADMET prediction that outperforms state-of-the-art baselines across 13 benchmarks and 21 tasks, achieving >75% predictions within 3-fold error on regression tasks and >50% within 2-fold on external HLMC predictions. It was validated on three preclinical candidates with HLMC within 2.5-fold and 73.3% correct CYP450 classification.
Predicting the absorption, distribution, metabolism, excretion and toxicity (ADMET) properties of small molecules remains a major challenge in drug discovery. Here, we present MEGA-CL, a foundation graph neural network framework for universal molecular ADMET prediction. MEGA-CL integrates self-supervised contrastive learning with a multi-head external attention mechanism and an enhanced message-passing architecture, enabling simultaneous modeling of local chemical substructures and global inter-graph relationships while mitigating over-smoothing effects commonly observed in deep graph networks. Across 13 benchmark datasets and 21 downstream ADMET tasks, MEGA-CL consistently outperforms state-of-the-art baseline models. In particular, the framework demonstrates robust performance on challenging regression tasks, including clearance (CL) and steady-state volume of distribution (VDss), while maintaining strong generalization ability in independent external validation. Clinically relevant predictive accuracy was achieved, with more than 75% of predictions falling within a 3-fold error range. In an external evaluation on 18 novel compounds derived from recently approved FDA drugs, over 50% of human liver microsome clearance (HLMC) predictions were within a 2-fold error range. To further assess its practical applicability, MEGA-CL was prospectively evaluated on three preclinical drug candidates using in vitro hepatic microsomal metabolism assays and CYP450 inhibition assays guided by model predictions. The predicted HLMC values for all candidates were within 2.5-fold of the experimentally measured values, and 73.3% of CYP450 inhibition endpoints (11/15) were correctly classified. These results demonstrate the potential of MEGA-CL as a generalizable framework for accelerating in silico ADMET evaluation and early-stage drug candidate optimization.