Probabilistic Contrastive Pretraining Framework Improves ADME Property Prediction for Drug Discovery
Researchers have developed a molecular graph-transformer pretraining method that combines contrastive learning with chemistry-specific self-supervision to predict ADME (absorption, distribution, metabolism, excretion) properties in drug molecules. The approach treats reconstruction, contrastive discrimination, and chemistry tasks as unified probabilistic objectives rather than separately weighted components. The method shows 7.6-9.9% improvements over baseline approaches across multiple datasets, addressing a critical challenge in drug discovery where ADME prediction is noisy and data-limited.
The study proposes Contrastive KERMT, a pretraining framework that encodes molecular graphs into latent variables and reconstructs SMILES strings while optimizing contrastive mutual information objectives. Rather than using separately tuned loss weights for different tasks, the authors formulate reconstruction, contrastive discrimination, and chemistry-specific supervision as unit-weighted log-probability factors in a single probabilistic objective. For fine-tuning, they introduce a multi-task GNN readout architecture with task-specific heads that preserves shared representation learning while mitigating negative transfer. Testing across three datasets (Biogen, ExpansionRX, and ChEMBL-MT) shows consistent improvements of 7.6-9.9% over the KERMT baseline on significantly-improved endpoints. The framework also benefits from including ADME-adjacent molecules in pretraining, and the contrastive component produces chemically meaningful latent neighborhoods.
What's missing
The study does not discuss computational costs, training time, or scalability compared to baseline methods. Limitations regarding the size and diversity of the datasets used, generalization to novel chemical scaffolds, and applicability to other molecular property prediction tasks beyond ADME are not explicitly addressed in the abstract.
What different sources said
- arXiv cs.LGCenter
Probabilistic Contrastive Pretraining for Multi-task ADME Property Prediction
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.