Enhancing virtual screening of cystathionine β-synthase inhibitors: benchmarking target-specific machine-learning scoring functions against state-of-the-art AI docking and co-folding approaches.

Cystathionine β-synthase (CBS) has emerged as an important therapeutic target implicated in cancer and Down syndrome, yet the discovery of selective CBS inhibitors remains challenging due to limited structural diversity of known ligands and the scarcity of target-focused virtual screening (VS) benchmarks. In this study, we present the first comprehensive evaluation of CBS-specific artificial intelligence (AI) models for structure-based VS, supported by a carefully curated and up-to-date data set of experimentally validated CBS inhibitors, true inactives and decoys. Using this data set, we developed CBS-specific binary classification models trained on docking-derived features and evaluated them in a rigorous five-fold cross-validation framework that employed similarity-controlled splits, ensuring structural independence between training and test sets while minimizing intra-fold class bias. Predictive performance was assessed primarily using the normalized enrichment factor of true actives at 1% (NEF1%) alongside balanced accuracy. Across all folds, the CBS-specific AI models demonstrated consistently high screening power and robust classification performance. Importantly, we benchmarked their performance against a diverse panel of 16 established VS pipelines, including widely adopted docking-based scoring schemes, modern deep-learning (DL) docking tools, and recent co-folding approaches for protein-ligand binding prediction. The CBS-specific AI classifiers substantially outperformed these state-of-the-art (SOTA) methods in early enrichment (NEF1% = 0.764 ± 0.191), highlighting the advantage of target-focused training for VS tasks. Our results further reveal that generic DL docking and co-folding approaches struggle to achieve reliable screening performance when applied to targets under-represented in or entirely absent from their training data, as appears to be the case for CBS. This finding underscores a key limitation of broadly trained foundation-style models in prospective drug discovery campaigns involving less-studied proteins. Overall, this work reinforces the benefits of target-specific AI scoring models tailored to individual proteins, highlights the value of high-quality curated data sets for such efforts, and provides practical guidance for selecting VS strategies in case of data-limited targets. To promote transparency and reproducibility, all input and output files, curated data sets and source code are freely available via GitHub and Zenodo.