Protein backbone diffusion models undersample loop-rich and alpha-beta domains and functional motifs

Summary

Protein backbone design models that use diffusion undersample alpha/beta topologies and loop regions, particular those found in functional motifs found in enzymes, and oversample helical domains not found in CATH (1). This includes designable sheet-rich domains like immunoglobulins and is consistent across all diffusion models tested.

Figures

Ref (1)

See also

• Diffusion-based protein design methods undersample structural diversity in specific topologies
• Protein structure prediction methods predict idealized secondary structures in de novo proteins

1.

Lu T, Liu M, Chen Y, Kim J, Huang P-S. Assessing generative model coverage of protein structures with SHAPES. Cell Systems. 2025;16(8):101347. Available from: https://doi.org/10.1016/j.cels.2025.101347