Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling

2026-05-31

AI Summary

Existing biomolecular co-design methods rely on fixed synchronous coupling, which often leads to high variance in supervisory signals and inconsistent intermediate states, thereby compromising cross-modal consistency. This work proposes GeoCoupling, a novel framework that, for the first time, systematically models the temporal coupling degrees of freedom in multimodal generation, transcending conventional synchronization constraints. By introducing an intrinsic geodesic coupling mechanism—combined with a structure-aware temporal alignment strategy and optimized heterogeneous diffusion processes—the method enables coherent joint generation of sequences and structures. Evaluated on structure-based drug design and unconditional protein generation tasks, GeoCoupling significantly outperforms both synchronous and randomly coupled baselines, yielding molecules with superior physical plausibility and enhanced diversity.

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.

Problem

• multimodal co-design
• temporal coupling
• biomolecular generation
• modality consistency
• generative modeling

Innovation

• multimodal co-design
• temporal coupling
• geodesic coupling
• generative modeling
• biomolecular design