Cheminformatics

Few-step Cofolding with All-Atom Flow Maps

Gianluca Scarpellini, Joey Bose

Featured June 22, 2026

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new method called D E CAF makes predicting how proteins and drugs fit together much faster by using flow maps to generate accurate 3D structures in just a few steps, instead of many.

In depth

The D E CAF framework distills computationally expensive all-atom diffusion models into flow maps, enabling the generation of high-quality 3D biomolecular structures in only a few inference steps. This acceleration is achieved through a novel

σ

-space reparameterization and an SE(3) rigid alignment loss, which also facilitates efficient reward-guided search for physically valid structures.

Key Takeaways

1
D E CAF significantly accelerates all-atom biomolecular structure prediction by distilling diffusion models into flow maps, reducing Neural Function Evaluations (NFEs) by 5-20x.
2
The framework introduces a $\sigma$-space reparameterization for stable training and a denoiser-based flow map parametrization that crucially enables SE(3) rigid alignment of ground truth structures.
3
D E CAF-SEARCH leverages the flow map's lookahead capability for higher-fidelity reward estimation, enabling efficient inference-time search to generate physically valid and reward-optimized structures.

Conceptual Flow

HIGH LEVEL

Methodology: How D E CAF Works

D E CAF learns a direct 'jump' from noisy protein-drug shapes to clean ones, making predictions super fast, and can even guess how good a shape will be before it's finished.

Noisy Protein-Drug Shape

Learn Direct Jump

Clean Protein-Drug Shape

Guess Quality

Results: What D E CAF Achieves

D E CAF creates accurate protein-drug shapes much quicker than old methods, using way less computer power, which helps scientists find new medicines faster.

Slow Old Method

Lots of Computer Power

D E CAF Improvement

Fast New Method

Less Computer Power

Accurate Shapes