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.
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.
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.
Kai Hodžić, Gustav Šír· ACM Transactions on Intellig...· 0 citations
The SBMR-CNN model demonstrates highly competitive accuracy, outperforming the CM, Uni-Mol+, and MPNN-2D benchmarks, while closely approaching the performance of the more computationally intensive MPNN-3D and SOAP descriptors, as well as the RF-MF model.
Abdulaziz W. Alherz, C. Tezak, Mohammed S. Alhajeri· Industrial & Engineering...· 0 citations
Graph Autoencoder Uncovering Descriptive Information (GAUDI), an unsupervised graph deep learning framework designed to capture both local details and global structure, is introduced, providing new insights into emergent phenomena across diverse scientific domains.
For many years, graph representation learning plays a pivotal role in bioinformatics and cheminformatics; as a result, supporting a wide range of tasks such as drug discovery, toxicity prediction, and compound–protein interaction analysis. However, existing approaches often focus solely on either sequential molecular fingerprints or graph‐based structural features, which limit their ability to capture both local chemical substructures and global molecular topology. To address this issue, we propose MM2Vec, a novel multi‐viewed molecular representation learning framework that integrates local rich‐feature embedding with graph neural network (GNN)‐based structural learning. Specifically, each molecular graph is first processed through an MLP‐based embedding layer that encodes sub‐structural fingerprint information extracted from radius‐based subgraphs, capturing fine‐grained chemical and physiochemical features. Simultaneously, a multi‐layered GNN encoder learns topological relationships from the molecular graph structure; therefore, focusing more on geometric and relational information among atoms. The outputs from both embedding branches are then fused using a learnable linear mechanism to produce unified, high‐quality molecular embeddings in a shared latent space. These fused representations are used to drive task‐specific prediction layers for addressing various learning objectives. We validate the proposed MM2Vec model on multiple graph learning tasks, including drug‐induced liver injury (DILI) classification and lethal dose (LD) molecular regression problems. Experimental results show that MM2Vec consistently outperforms classical machine learning (ML)‐based models and recent state‐of‐the‐art deep learning (DL)/GNN‐based methods in terms of accuracy, robustness, and generalization. Our findings in this highlight the importance of combining both sub‐structural and graph‐structural perspectives and demonstrate the versatility and effectiveness of our MM2Vec model for a wide range of molecular analysis tasks.
Phu Pham· International Conference on...· 0 citations
Evaluations on eight MoleculeNet datasets show that MSMPP significantly outperforms state-of-the-art models, demonstrating its effectiveness in integrating multi-view intra-molecular features, inter-molecular features and cross-task information.
Jiongfeng Chen, Yulian Ding, Yan Yan et al.· IEEE journal of biomedical a...· 0 citations
Learning effective molecular representations is crucial for accurate property prediction in AI-aided drug discovery. However, most existing molecular pre-training methods are still primarily based on 2D topological graphs, limiting their ability to exploit 3D geometric information. Moreover, methods that do incorporate 3D geometry often do not distinguish between the roles of atom-centered and bond-centered representations. To address these limitations, we propose GDGraph, a geometryenhanced dual-view framework for molecular representation learning. GDGraph models molecular geometry from two complementary structural perspectives: an atom view for capturing global spatial dependencies and a bond view for modeling local geometric patterns. To support this dual-view design, we introduce a multi-scale geometric feature encoding scheme and a view-specific geometry-aware learning strategy, enabling each view to focus on the geometric dependencies it is best suited to capture. Extensive experiments demonstrate that GDGraph achieves strong and stable performance on molecular property prediction benchmarks, and effectively predicts geometrysensitive quantum chemical properties on the QM9 dataset.
Yu Liu, Jonathan D. Hirst, Jianfeng Ren et al.· Annual International Compute...· 0 citations