Probabilistic Contrastive Pretraining for Multi-task ADME Property Prediction

2026-06-09

AI Summary

This work addresses the challenges in ADME property prediction—namely data noise, inter-task dependencies, and limited sample availability—by introducing a molecular graph Transformer-based pretraining framework. The approach unifies chemically informed self-supervision (e.g., SMILES reconstruction), contrastive mutual information maximization (cMIM), and a multi-task GNN readout mechanism under a single probabilistic latent-variable objective, enabling joint optimization across reconstruction, discrimination, and downstream tasks. Crucially, diverse self-supervised signals are modeled as equally weighted probabilistic factors, eliminating manual hyperparameter tuning, while task-specific MLP heads mitigate negative transfer and capture complex nonlinear task relationships. The method outperforms baselines by 7.6%, 9.9%, and 9.5% on Biogen, ExpansionRX, and ChEMBL-MT benchmarks, respectively; incorporating ADME-relevant molecules further enhances transferability, and ablation studies confirm that the proposed components effectively enrich chemical semantic neighborhood representations.

Abstract

Accurate prediction of absorption, distribution, metabolism, and excretion (ADME) properties is critical to drug discovery, but remains challenging because ADME endpoints are noisy, interdependent, and often data-limited. We propose a molecular graph-transformer pretraining framework that combines chemistry-specific self-supervision with contrastive mutual information machine learning (cMIM). Our method encodes molecular graphs into latent variables, reconstructs SMILES strings from the graph-derived latent codes, and augments the contrastive objective with domain-specific self-supervised chemistry tasks. Rather than treating these tasks as auxiliary regularizers with separately tuned loss weights, we formulate reconstruction, contrastive discrimination, and chemistry-specific supervision as unit-weighted log-probability factors in a single probabilistic latent-variable objective. For fine-tuning, we propose a multi-task GNN readout architecture with task-specific multilayer perceptron heads, preserving shared representation learning while mitigating negative transfer and improving the modeling of heterogeneous, nonlinear task relationships. Across Biogen, ExpansionRX, and ChEMBL-MT, the resulting Contrastive KERMT pretraining improves over the KERMT baseline by 7.6%, 9.9%, and 9.5% respectively (averaged over significantly-improved endpoints). Adding ADME-adjacent molecules to the pretraining corpus further improves transfer, and the contrastive component sharpens chemically meaningful latent neighborhoods.

Problem

• ADME prediction
• multi-task learning
• data-limited
• noisy endpoints
• interdependent properties

Innovation

• contrastive pretraining
• probabilistic latent-variable model
• multi-task GNN
• chemistry-specific self-supervision
• ADME property prediction