DCFold: Efficient Protein Structure Generation with Single Forward Pass
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.
Related
- Efficient generation of epitope-targeted antibodies with Germinal
- Human cytomegalovirus microRNAs target the Wnt signaling protein Dishevelled 3 to promote efficient reactivation in CD34+ hematopoietic progenitor cells
- TopoFlow: Evolutionarily Conditioned Flow Matching for Protein Conformational Ensemble Generation
- DeepD enables efficient genome modeling and controllable long-context sequence generation across domains of life
- N1-Methylpseudouridine-Modified mRNA-Transfected Dendritic Cells Enable Efficient Generation of Functional Tumor Antigen-Specific Cytotoxic T Lymphocytes
BibTeX
@misc{zhang2026dcfold,
title = {{DCFold: Efficient Protein Structure Generation with Single Forward Pass}},
author = {Zhe Zhang and Yuanning Feng and Yuxuan Song and Keyue Qiu and Hao Zhou and Wei-Ying Ma},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2605.17899},
eprint = {2605.17899},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.17899},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - DCFold: Efficient Protein Structure Generation with Single Forward Pass
AU - Zhe Zhang
AU - Yuanning Feng
AU - Yuxuan Song
AU - Keyue Qiu
AU - Hao Zhou
AU - Wei-Ying Ma
PY - 2026
DA - 2026/09/26
PB - arXiv
DO - 10.48550/arxiv.2605.17899
UR - https://arxiv.org/abs/2605.17899
AB - 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.
ER -