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

2026-07-06

AI Summary

This work addresses the challenge in molecular property prediction that single-modality models struggle to simultaneously capture long-range topological dependencies and macroscopic physicochemical information. To overcome this limitation, the authors propose a tri-branch modular fusion network that synergistically integrates 3D geometric representations (via SchNet), SMILES-based topological syntax (via ChemBERTa), and explicit macroscopic descriptors (via a Deep & Cross Network) within a parameter-efficient late-fusion architecture to construct a unified latent space. This approach uniquely combines orthogonal macroscopic and topological information streams, providing an O(1) physical shortcut that circumvents the oversmoothing and arithmetic limitations inherent in local message-passing graph neural networks. On the QM9 benchmark, the model achieves a state-of-the-art MAE of 0.0207 eV for U₀^atom prediction with fewer than one million parameters—representing a 20.6% error reduction over purely geometric baselines and surpassing the sub-chemical accuracy threshold.

Abstract

Molecular property prediction often relies on isolated data modalities, where continuous 3D graph neural networks (GNNs) struggle to efficiently capture long-range topological dependencies and exact macroscopic heuristics. In this work, we introduce a parameter-efficient Tri-Branch Modular Fusion Neural Network that synthesizes three orthogonal modalities: 3D spatial geometry (SchNet), discrete topological grammar (SMILES via ChemBERTa), and explicit macroscopic physicochemical descriptors (Deep & Cross Network). By bypassing standard scalar readouts and employing a shared late-fusion architecture, the framework establishes a mathematically rigorous multimodal latent space that effectively resolves the arithmetic and oversmoothing limitations of local message passing. We evaluate the proposed architecture on the QM9 benchmark, targeting the extensive thermodynamic property of atomization energy at 0 K ($U_0^{\mathrm{atom}}$). Through systematic combinatorial ablation and latent bottleneck optimization ($d_e=64$), the tri-modal framework achieves a validation Mean Absolute Error (MAE) of 0.0207 eV. Operating with fewer than one million parameters, this architecture decisively surpasses the sub-chemical accuracy threshold and yields a substantial 20.6% error reduction over a strictly controlled geometric baseline. Ultimately, our findings demonstrate that integrating orthogonal macroscopic and topological data streams provides a synergistic, $\mathcal{O}(1)$ physical shortcut. This multimodal alignment offers a highly efficient alternative to brute-force parameter scaling, establishing a robust surrogate model for high-throughput virtual screening (HTVS) pipelines.

Problem

• molecular property prediction
• multimodal representation
• long-range dependencies
• topological grammar
• macroscopic descriptors

Innovation

• Multimodal Fusion
• Graph Neural Networks
• SMILES Embeddings
• Deep & Cross Network
• Molecular Property Prediction