Summary
In a blinded prospective benchmark, most computational rankings of antibodies within HCDR3 clusters underperformed random clone selection (1). For the 27F, 28F, and 47F clusters, 10.3%, 13.8%, and 9.8% of submitted predictions beat the affinity of the most abundant clone, compared with 39% of randomly selected clones. The 28F figure caption reports 5.2% for a filtered analysis rather than the text’s 13.8%. The only method to beat the baseline combined the competition model with an external dataset of about 3,000 SARS-CoV-2 affinity measurements.
See also
- NGS sequence abundance does not correlate with binding affinity
- CDRH3 is necessary but insufficient to guarantee binding to a particular epitope
1.
Erasmus MF, Bedinger D, Hopkins E, Ferguson G, Strickler J, Graff CP, et al. A blinded, prospective benchmark of in silico antibody discovery anchored to experimental affinity and developability. Nature Biotechnology. 2026; Available from: https://doi.org/10.1038/s41587-026-03238-6