Emyx: Fast and efficient all-atom protein generation

2026-06-12

AI Summary

This work addresses the challenges of high training costs and limited sampling diversity in current all-atom protein generative models, which struggle to simultaneously achieve catalytic site geometric precision and structural novelty. The authors propose a lightweight conditional flow-matching model that leverages sparse connectivity, compact conditional embeddings, and an architecturally simplified design to efficiently generate diverse and valid enzyme backbones under strict geometric constraints. Innovatively, they reparameterize flow-matching interpolation precisely within the EDM noise framework, enabling seamless integration with advanced diffusion sampling strategies without retraining. Evaluated on the AME enzyme design benchmark, their 140M-parameter model outperforms Proteína-Complexa and RFdiffusion3 across multiple metrics—including fold recovery, catalytic geometry accuracy, and structural novelty—while requiring only 682 GPU hours for training, approximately one-quarter of RFdiffusion3’s cost.

Abstract

Computational enzyme design requires generating proteins that scaffold catalytic residues and ligands, a task that demands both geometric accuracy and structural diversity from the underlying generative model. Current all-atom generators inherit expensive architectures from structure prediction, leading to high training costs and limited sample diversity. We argue that much of this complexity is unnecessary for generators, which condition on sparse geometric constraints rather than rich co-evolutionary signals. Emyx is a 140M-parameter conditional flow matching model that concentrates capacity within standard transformer blocks, replacing heavy embedding stacks with lightweight conditional representations and sparse connectivity. We additionally derive an exact reparametrisation of the flow matching interpolant into the EDM noise-level framework, bridging flow matching training efficiency with state-of-the-art sampling methods designed for diffusion models without retraining. Despite being the smallest model, Emyx outperforms both Proteína-Complexa and RFdiffusion3 against the AME enzyme design benchmark across success rate under strict evaluation requiring both global fold recovery and catalytic geometry accuracy, structural novelty, scaffold diversity, and geometric validity, while training in just $682$ GPU-hours, roughly $4\times$ less than RFdiffusion3.

Problem

• protein generation
• enzyme design
• all-atom modeling
• structural diversity
• geometric accuracy

Innovation

• conditional flow matching
• all-atom protein generation
• lightweight transformer architecture
• EDM reparameterization
• computational enzyme design