Accurate prediction of drug-target affinity (DTA) is central to drug discovery but remains challenging in low-data regimes where deep learning models generalize poorly. This study introduces AdaMBind, a meta-learning-based framework with adaptive task scheduling that substantially improves few-shot DTA prediction, enabling robust identification of high-affinity compounds and experimentally validated inhibitor discovery under stringent data constraints.
- Mengxuan Wan
- Yanpeng Zhao
- Xiaochen Bo
