HonestAffinity: Leak-Aware Evaluation of Protein and Pocket Priors for Binding Affinity Prediction
Junhao Wei 1 , Baili Lu 1,2 , Zhenhong Peng 2 , Wanyan Li 2 , Zhirong Huang 3 , Yanxiao Li 1 , Yifu Zhao 1 , Dexing Yao 1 , Haochen Li 1 , Xudong Ye 1 , Sio-Kei Im 4 , Yapeng Wang 1 , Xu Yang 1,* · 2026
Read HonestAffinity: Leak-Aware Evaluation of Protein and Pocket Priors for Binding Affinity Prediction by Junhao Wei 1 , Baili Lu 1,2 , Zhenhong Peng 2 , Wanyan Li 2 , Zhirong Huang 3 , Yanxiao Li 1 , Yifu Zhao 1 , Dexing Yao 1 , Haochen Li 1 , Xudong Ye 1 , Sio-Kei Im 4 , Yapeng Wang 1 , Xu Yang 1,* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-03422.txt.
Description
Wei and co-authors test how well simple inputs can predict protein-ligand binding strength. Sequence-based deep learning offers a cheap path versus structure-based scoring. But common test splits leak similar cases across training and test folds. That leak makes new model parts look better than they are. HonestAffinity is a compact predictor built to face that problem.
It uses frozen ESM-2 protein embeddings plus a learned marker for pocket positions. Three variants switch the pocket marker and the embeddings on or off. All train on 11,513 complexes in about three GPU hours. Tests cover a strict no-leak hold-out and the CASF-2016 benchmark. The pocket version wins on familiar splits. The no-ESM pocket version wins on every strict no-leak tier. Both parts help on easy splits yet hurt on strict ones. The authors urge paired reports of standard and leak-proof tests.
Catalog record
| CREATOR | Junhao Wei 1 , Baili Lu 1,2 , Zhenhong Peng 2 , Wanyan Li 2 , Zhirong Huang 3 , Yanxiao Li 1 , Yifu Zhao 1 , Dexing Yao 1 , Haochen Li 1 , Xudong Ye 1 , Sio-Kei Im 4 , Yapeng Wang 1 , Xu Yang 1,* |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.03422