AI-Driven Drug-Target Interaction Prediction: From Data Representation to Model Design — A Systematic Review

The review frames the evolution of DTI prediction as a shift from docking, similarity-based inference, and hand-crafted features toward data-driven representation learning and interaction modeling. Compared with prior surveys focused on a single method family, it ties this shift to the growth of biomedical data and advances in AI as the organizing thread of the whole review. A survey-level claim based on synthesis of the field's direction; no quantitative comparison is provided in the text.

The review places classical molecular binding theories (lock-and-key, induced fit, conformational selection) at the starting point of its methodological discussion, emphasizing the transition from static matching to dynamic interaction. It juxtaposes mechanism-level binding theory with computational modeling approaches, providing a conceptual basis for understanding model design motivations. A conceptual synthesis citing classical theoretical frameworks rather than new experimental data.

The review summarizes major DTI task settings: binary interaction classification, binding affinity regression, and multitask prediction with uncertainty assessment. It lists uncertainty assessment alongside multitask prediction as part of the task settings, extending the traditional split into classification and regression only. A taxonomy-level summary; no specific benchmark numbers are reported.

The review compares representative approaches across orthogonal dimensions including input representation, encoder architecture, interaction-modeling mechanism, representation learning and pretraining, learning objective, prediction output, data acquisition or optimization strategy, and generalization setting. It offers a multidimensional comparison framework that supports cross-family comparison rather than listing methods by time or a single technical route. The comparison framework is derived from synthesizing representative approaches; no unified experimental head-to-head evaluation is provided.

The article is positioned as a systematic reference for readers who want an overall picture of the methodological landscape, task settings, and representation choices in AI-driven DTI prediction, including algorithm designers, mechanism researchers, and drug discovery practitioners; its organization around binding theory, task settings, data and representations, model design dimensions, applications, and challenges makes it a suitable starting point for entering the field or planning a research route.

This reading is a fast parse at the abstract level and does not include figures, methodological details, or the reference list from the full text, so it cannot present specific performance comparisons, dataset sizes, or experimental setups for the representative approaches; readers who need to make method-selection decisions should consult the original figures and cited literature. In addition, the challenges mentioned in the text, such as data distribution shifts, dynamic protein conformational variability, and insufficient experimental validation, require further confirmation against the original text to understand their concrete manifestations and proposed responses.