XIMP: Cross Graph Inter-Message Passing for Molecular Property Prediction

TL;DR

XIMP introduces a novel cross-graph message passing framework that integrates multiple molecular graph abstractions, improving property prediction accuracy especially in low-data regimes by leveraging chemical domain knowledge.

Contribution

The paper presents XIMP, a flexible message passing method across multiple graph abstractions, surpassing prior single-abstraction or non-iterative approaches in molecular property prediction.

Findings

XIMP outperforms state-of-the-art models on ten molecular property tasks.

XIMP effectively leverages chemical abstractions to improve generalization.

XIMP enhances low-data learning through interpretable graph representations.

Abstract

Accurate molecular property prediction is central to drug discovery, yet graph neural networks often underperform in data-scarce regimes and fail to surpass traditional fingerprints. We introduce cross-graph inter-message passing (XIMP), which performs message passing both within and across multiple related graph representations. For small molecules, we combine the molecular graph with scaffold-aware junction trees and pharmacophore-encoding extended reduced graphs, integrating complementary abstractions. While prior work is either limited to a single abstraction or non-iterative communication across graphs, XIMP supports an arbitrary number of abstractions and both direct and indirect communication between them in each layer. Across ten diverse molecular property prediction tasks, XIMP outperforms state-of-the-art baselines in most cases, leveraging interpretable abstractions as an inductive bias that guides learning toward established chemical concepts, enhancing generalization in low-data settings.