Abstract
<jats:p>Computational drug discovery is often narrated as a series of revolutions, each new architecture rendering the last obsolete. This review instead reads six decades of the field, from Hansch and Fujita's 1964 QSAR framework to current diffusion and foundation models, as a single methodological lineage in which later tools extend and absorb earlier ones rather than replace them outright. Five major methodological transitions are examined against the relationship between data availability and method adoption, yielding a specific pattern rather than a general orientation: three transitions, the rise of structure-based design, the shift to broadly applicable machine-learning QSAR, and the current wave of structure-conditioned generative methods, were gated by prior or concurrent expansion of data infrastructure (the Protein Data Bank, ChEMBL, and the AlphaFold Database respectively), while two, the shift from linear regression to non-linear ensemble methods and the 2012 deep learning inflection, were driven primarily by architectural innovation with data availability playing a secondary role. Antimicrobial resistance is used throughout as a stress test, since its combination of target novelty, data sparsity, and evolving resistance mechanisms exposes where each generation of methods has, and has not, generalised, and two documented successes, halicin and abaucin, are examined in detail alongside the downstream synthetic and toxicological attrition that followed their identification. Data debt, meaning unrepresentative, sparse, and inequitably distributed training data, particularly for infectious disease and low-resource contexts, emerges as a persistent constraint that architectural progress alone has not resolved. Seven open problems are identified, most concerning data, validation, and governance rather than model design, and a specific, testable comparison is proposed by which the relative contribution of data scale versus architectural advance could be assessed directly using existing AMR data consortia.</jats:p>