Emyx: Fast and efficient all-atom protein generation

Emyx is a new 140M-parameter conditional flow matching model designed for efficient all-atom protein generation, emphasizing geometric accuracy and structural diversity. It utilizes standard transformer blocks with lightweight conditional representations, significantly reducing training costs to 682 GPU-hours—approximately four times less than RFdiffusion3. Emyx demonstrates superior performance on the AME enzyme design benchmark, outperforming larger models like Proteína-Complexa and RFdiffusion3 in terms of strict evaluation metrics, making it a valuable tool for practitioners in computational enzyme design.