MultiRankDTA: A Hybrid Ranking Framework for Robust and Generalizable Drug-Target Affinity Prediction

Abstract Predicting drug-target affinity (DTA) is essential for screening candidate compounds in early-stage drug discovery. However, most existing computational methods formulate DTA prediction as a regression task, leading to an objective mismatch with the practical need to rank and prioritize candidate compounds based on their relative binding strengths. Moreover, existing ranking-based methods often adopt a fixed ranking objective or scoring function, which may limit their flexibility when they are applied to diverse drug-target interaction scenarios. To address these challenges, we propose MultiRankDTA, a novel framework that formulates DTA prediction as a multilevel ranking task to better align with the practical goal of drug screening. By integrating pointwise, pairwise, and listwise ranking strategies through a mixture-of-experts (MoE) module, MultiRankDTA enables dynamic and adaptive compound prioritization. Furthermore, we introduce a fusion network to reduce data set-specific biases by leveraging auxiliary supervision from multiple sources. Extensive experiments demonstrate that MultiRankDTA achieves superior performance compared to state-of-the-art computational methods across multiple benchmark data sets and delivers strong results on an independent benchmark for virtual screening, indicating its potential for generalization and practical utility. The source code can be obtained at https://github.com/CSUBioGroup/MultiRankDTA.