DCFold: Efficient Protein Structure Generation with Single Forward Pass

TL;DR

DCFold is a novel single-pass generative model for protein structure prediction that matches AlphaFold3's accuracy while being 15 times faster, enabling more practical downstream applications.

Contribution

The paper introduces DCFold, a single-step model with a Dual Consistency training framework and TGM scheduler, significantly reducing inference time without sacrificing accuracy.

Findings

DCFold achieves AlphaFold3-level accuracy.

Inference time is reduced by 15x with DCFold.

Validated effectiveness on structure prediction and binder design.

Abstract

AlphaFold3 introduces a diffusion-based architecture that elevates protein structure prediction to all-atom resolution with improved accuracy. This state-of-the-art performance has established AlphaFold3 as a foundation model for diverse generation and design tasks. However, its iterative design substantially increases inference time, limiting practical deployment in downstream settings such as virtual screening and protein design. We propose DCFold, a single-step generative model that attains AlphaFold3-level accuracy. Our Dual Consistency training framework, which incorporates a novel Temporal Geodesic Matching (TGM) scheduler, enables DCFold to achieve a 15x acceleration in inference while maintaining predictive fidelity. We validate its effectiveness across both structure prediction and binder design benchmarks.