Research map / Label-efficient and robust learning / Self-supervised and contrastive

PRISM: Prior Relational Information for Self-supervised Modeling to Enhance Solubility OOD Generalization

NeurIPS 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes PRISM, a self-supervised pretraining approach that injects prior relational information into solute-solvent representations to improve out-of-distribution solubility prediction.

Problem

Solubility prediction must generalize out-of-distribution to unseen solutes, solvents, and solvent mixtures, where supervised models suffer from data scarcity and existing self-supervised pretraining does not explicitly learn chemically meaningful solute-solvent interactions.

Approach

Explicitly encodes prior relational information about intermolecular solute-solvent interactions into a self-supervised pretraining objective, rather than relying on generic self-supervised pretraining.

Closest papers by meaning

• Relational Invariant Learning for Robust Solvation Free Energy Prediction (ICML 2025 · similarity 0.89)
• SC$^3$: A Multi-Solvent Solubility Challenge and Benchmark (NeurIPS 2026 · similarity 0.87)
• CheMixHub: Datasets and Benchmarks for Chemical Mixture Property Prediction (NeurIPS 2025 · similarity 0.86)
• From Molecules to Materials: Pre-training Large Generalizable Models for Atomic Property Prediction (ICLR 2024 · similarity 0.85)
• Multimodal Molecular Pretraining via Modality Blending (ICLR 2024 · similarity 0.85)
• ChemPile: A 250 GB Diverse and Curated Dataset for Chemical Foundation Models (NeurIPS 2025 · similarity 0.85)