Known Unknowns: Out-of-Distribution Property Prediction in Materials and Molecules

2025-02-09

AI Summary

This work addresses the zero-shot out-of-distribution (OOD) prediction challenge in materials and molecular design—i.e., reliably predicting high-value performance properties beyond the training data distribution. Conventional inductive models are fundamentally limited by the scope of their training objectives. To overcome this, we propose the first analogical-reasoning-driven transductive OOD prediction paradigm, jointly optimizing chemical representation learning, graph neural networks, and OOD classification. Unlike standard approaches, our method makes no assumptions about the label distribution of target properties; instead, it achieves cross-distribution generalization by explicitly modeling analogical relationships among samples. Experiments demonstrate substantial improvements: on solid-materials OOD classification, true positive rate (TPR) increases 3× and precision 2×; on molecular property prediction, TPR improves 2.5× and precision 1.5×—all significantly surpassing non-transductive baselines.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Cognitive Modeling & Cognitive Systems: Analogy
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Discovery of high-performance materials and molecules requires identifying extremes with property values that fall outside the known distribution. Therefore, the ability to extrapolate to out-of-distribution (OOD) property values is critical for both solid-state materials and molecular design. Our objective is to train predictor models that extrapolate zero-shot to higher ranges than in the training data, given the chemical compositions of solids or molecular graphs and their property values. We propose using a transductive approach to OOD property prediction, achieving improvements in prediction accuracy. In particular, the True Positive Rate (TPR) of OOD classification of materials and molecules improved by 3x and 2.5x, respectively, and precision improved by 2x and 1.5x compared to non-transductive baselines. Our method leverages analogical input-target relations in the training and test sets, enabling generalization beyond the training target support, and can be applied to any other material and molecular tasks.

Problem

• Extrapolate to out-of-distribution property values
• Improve prediction accuracy for materials and molecules
• Leverage analogical input-target relations for generalization

Innovation

• Transductive approach for OOD prediction
• Zero-shot extrapolation beyond training data
• Analogic input-target relations for generalization