Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling

2026/06/01 by Keyue Qiu, Xintong Wang, Zhilong Zhang +2 · 1 voice
Biochemistry, Genetics and Molecular Biology · Computer Science · #q-bio.BM #cs.AI

paper · pdf

Abstract

Biomolecules such as proteins and small-molecule ligands play a central role in biological systems, arising from the tight interplay between sequence and three-dimensional structure. Recent generative models for biomolecular co-design aim to capture this interplay by jointly modeling coupled modalities. However, existing approaches largely adopt a parallel execution of marginal generative processes, implicitly enforcing fixed synchronous coupling. We argue that a critical but overlooked degree of freedom lies in how these marginal processes are temporally coupled during training and generation, where inappropriate coupling can introduce high-variance supervision and inconsistent intermediate states, affecting modality consistency. To address this, we introduce GeoCoupling, a systematic framework that optimizes for temporal couplings between heterogeneous modalities. Empirical results across structure-based drug design and unconditional protein design demonstrate the learned couplings consistently outperform synchronous and randomly coupled baselines, yielding biomolecules with improved physical validity and diversity.

Citations

• Multi-domain Distribution Learning for De Novo Drug Design
• La-Proteina: Atomistic Protein Generation via Partially Latent Flow Matching
• Piloting Structure-Based Drug Design via Modality-Specific Optimal Schedule
• AlphaFold Protein Structure Database and 3D-Beacons: New Data and Capabilities
• DPLM-2: A Multimodal Diffusion Protein Language Model
• ProteinBench: A Holistic Evaluation of Protein Foundation Models
• Accurate structure prediction of biomolecular interactions with AlphaFold 3
• MolCRAFT: Structure-Based Drug Design in Continuous Parameter Space
• Diffusion Language Models Are Versatile Protein Learners
• DecompDiff: Diffusion Models with Decomposed Priors for Structure-Based Drug Design
• Generative Flows on Discrete State-Spaces: Enabling Multimodal Flows with Applications to Protein Co-Design
• SE(3)-Stochastic Flow Matching for Protein Backbone Generation
• Bayesian Flow Networks
• PoseBusters: AI-based docking methods fail to generate physically valid poses or generalise to novel sequences
• De novo design of protein structure and function with RFdiffusion
• 3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction
• Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation
• SE(3) diffusion model with application to protein backbone generation
• Evolutionary-scale prediction of atomic-level protein structure with a language model
• Structure-based Drug Design with Equivariant Diffusion Models
• Flow Matching for Generative Modeling
• Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow
• Learning inverse folding from millions of predicted structures
• ProGen2: Exploring the Boundaries of Protein Language Models
• Robust deep learning–based protein sequence design using ProteinMPNN
• Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets
• A 3D Generative Model for Structure-Based Drug Design
• Fast and accurate protein structure search with Foldseek
• Highly accurate protein structure prediction with AlphaFold
• AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings
• ProGen: Language Modeling for Protein Generation

Discussions

• Keyue Qiu, Xintong Wang, Zhilong Zhang, Hao Zhou, Wei-Ying Ma: Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling https://arxiv.org/abs/2606.01628 https://arxiv.org/pdf/26 [bsky, 0 points, 0 comments]

Related

• BInD: Bond and Interaction-generating Diffusion Model for Multi-objective Structure-based Drug Design
• UniGenX: a unified generative foundation model that couples sequence, structure and function to accelerate scientific design across proteins, molecules and materials
• THFlow: A Temporally Hierarchical Flow Matching Framework for 3D Peptide Design
• Progressive Multi-Modality Learning for Inverse Protein Folding
• Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling