Summary

In epitope-guided de novo antibody design, applying strong guidance early for epitope targeting and annealing it to zero can preserve high targeting fidelity while reducing steric clashes and biophysical distortions (1).

Details

Static classifier-free guidance in TiDE-Ab gave heavy-chain epitope recall of 0.934 and 43.35 clashes per design. Combining cosine annealing with truncation at preserved recall at 0.935, increased epitope precision from 0.777 to 0.889, and reduced clashes to 8.14. These results are in silico and await experimental validation.

Figures

Guidance configurationHC-Prec (↑)HC-Rec (↑)F. Dev (↓)Binding mode (↑)Clashes (↓)
Static CFG0.777 ± 0.0440.934 ± 0.0410.2238 ± 0.07424.25 ± 1.2443.35 ± 35.17
CFG w/Early Truncation0.874 ± 0.0420.937 ± 0.0410.2103 ± 0.06814.27 ± 1.3010.62 ± 14.59
CFG w/Cosine Annealing0.876 ± 0.0430.939 ± 0.0420.2100 ± 0.06744.15 ± 1.179.76 ± 13.84
TiDE-Ab (Final)0.889 ± 0.0420.935 ± 0.0410.2083 ± 0.06774.42 ± 1.178.14 ± 11.93

Ref (1); TiDE-Ab uses both cosine annealing and early truncation

See also

1.
Kim Y, Baek M. De novo epitope-specific antibody design via time-dependent guidance. Bioinformatics. 2026;42(Supplement_2):btag455. Available from: https://doi.org/10.1093/bioinformatics/btag455