Skip to content
Open access

Deep Learning for Molecular Property Prediction with Logical Background Knowledge

Jul 2026 · ACM Transactions on Intelligent Systems and Technology · 0 citations · 26 references

TL;DR

This work proposes an enhanced learning framework that deeply integrates structured logical knowledge within GNN models, and demonstrates that incorporating domain-specific relational knowledge leads to better generalization and robustness compared to standard GNNs.

Abstract

Graph Neural Networks (GNNs) have emerged as powerful tools for molecular property prediction by leveraging graph-structured data representations of chemical compounds. However, standard GNN architectures are completely generic and do not allow for explicitly encoding domain-specific chemical knowledge, such as functional groups, molecular motifs, and atomic interactions, which can be critical for accurate predictions, especially in low-data regimes. In this work, we propose an enhanced learning framework that deeply integrates structured logical knowledge within GNN models. Particularly, by formulating the underlying message-passing GNN mechanisms as logic-driven transformations, we bridge the gap between symbolic domain knowledge and deep learning, improving predictive accuracy and explainability. Empirical results on multiple benchmark molecular datasets demonstrate that incorporating domain-specific relational knowledge leads to better generalization and robustness compared to standard GNNs.

Read PDF

Similar papers

Preprint Jul 2026

Model Agnostic Graph Prompt Learning for Crystal Property Prediction

A soft prompt learning framework that captures latent features essential for property prediction, which are not explicitly provided to the GNN, and which seamlessly integrates with any existing GNN encoder is proposed.

Shrimon Mukherjee, Kishalay Das, Partha Basuchowdhuri et al. · 0 citations
Dataset Open access Jul 2026

VitaGraph: building a knowledge graph for biologically relevant learning tasks

VitaGraph is presented, a comprehensive multi-purpose biological knowledge graph built by integrating and refining multiple public datasets and enabling benchmarking of graph-based models and offering the opportunity to tackle tasks such as drug repurposing, PPI prediction, and side-effect prediction, among others.

Francesco Madeddu, Lucia Testa, Gianluca De Carlo et al. · 0 citations
Aug 2026

Dual-Attention Multimodal Framework for Molecular Property Prediction.

A novel Dual-Attention Multimodal framework for Graphs and Sequence-based representations, so-called DAM-GS, which provides a promising solution for molecular property prediction with broad applications in drug discovery and computational molecular science.

Bay Van Nguyen, Vinh Truong, Ha Duong Thi Hong et al. · 0 citations
Preprint Jul 2026

Multimodal Molecular Representation Learning with Graph Neural Networks, Deep&Cross Networks, and SMILES Embeddings

This work introduces a parameter-efficient Tri-Branch Modular Fusion Neural Network that synthesizes three orthogonal modalities: 3D spatial geometry, discrete topological grammar, and explicit macroscopic physicochemical descriptors that offers a highly efficient alternative to brute-force parameter scaling.

Qiwei Han, Chi Zhou, Ruo-Yuan Wang et al. · 0 citations
Preprint Aug 2026

Multi-Granular Rationale-Guided Molecular LLM for Property Prediction

This is the first method to expose GNN-derived attributions to an LLM as evidence for property prediction, and achieves the best overall results among generalist models and narrows the gap to specialist models tuned for each task.

Junwoo Park, Minyoung Shin, C. Lee et al. · 0 citations
Open access Aug 2026

Transfer Message Passing for Molecular Property and Function Prediction across Polymer and Protein Graphs

The prediction of molecular properties and biological functions across vastly different scales of macromolecules presents a formidable challenge in computational chemistry and bioinformatics. Traditional computational approaches often treat synthetic polymers and biological proteins as fundamentally distinct entities, applying specialized algorithms that fail to leverage the shared topological and chemical principles underlying both domains. This paper introduces a novel computational framework based on Transfer Message Passing, designed to unify the representation and predictive modeling of polymer and protein graphs. By conceptualizing both classes of macromolecules as complex, attributed graphs where nodes represent constituent functional units and edges denote chemical or spatial interactions, we establish a generalized topological space suitable for advanced graph neural networks. The proposed transfer learning mechanism dynamically adapts message passing operations learned from data-rich protein databases to infer complex physical and thermodynamic properties in specialized polymer datasets, mitigating the pervasive issue of data scarcity in polymer informatics. Extensive empirical evaluations demonstrate that our framework significantly outperforms domain-specific baseline models in predicting polymer bandgaps, glass transition temperatures, and protein enzymatic functions. Furthermore, detailed ablation studies reveal that the cross-domain attention mechanisms effectively align latent representations without compromising task-specific predictive accuracy. Ultimately, this research provides a robust theoretical foundation and a scalable computational tool for accelerated materials discovery and biomolecular engineering.

Richard Yat-Long Ma, Jessica Wing-Yan Lai · 0 citations