The abstract is read at the publisher; the summary is JClub's.
A noise-corrected, censoring-aware estimator shows cross-target antibody affinity transfer has a small ceiling, with regime-dependent budgets around 0.44–0.59 on real panels.
- Why it matters: Structure-based affinity generalization is expected to transfer across targets, but prior estimates are unclear and can be biased by measurement noise and censoring.
- What they did: The authors estimate a cross-target transfer ceiling, ρmax = √(T/(T+I)), separating transferable antibody-intrinsic signal (T) from non-transferable target-specific interaction (I), and validate it on synthetic ground truth and real neutralization/escape panels.
- The result: Under leakage-free antigen-cluster splits, cross-target ranking collapses to the budget, while within-target ranking improves (Spearman 0.55) and measure-then-rank recovers top binders 6–8× more efficiently, yet models do not exceed the ceiling.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.