Drug Design and Discovery: The Long Road From Hansch Analysis to Artificial Intelligence.

Abstract

The present article traces the long journey of the historical evolution and transformation of QSAR from its early days to its current role in drug discovery. It begins with the breakthrough created by the introduction of data analysis and highlights the major advances achieved throughout its development. Changes in molecular representation resulted in an explosion of molecular descriptors, while the rapid accumulation of large amounts of data demanded the application of elaborated statistical techniques. Today, machine learning and artificial intelligence have become integral components of pharmaceutical research, supporting every stage of the drug discovery pipeline, from target identification and hit discovery to lead optimization and the prediction of ADMET properties. At the same time, strict prerequisites have been established to ensure the development of QSAR models capable of reliable and robust predictions. Model validation, applicability domain, and model interpretability are fundamental principles of robust QSAR modeling. These principles are closely related to the distinction between local and global models, reflecting the trade-off between predictive performance, generalizability, and mechanistic interpretability. Finally, multi-objective optimization is discussed, as an approach to address the complex and often conflicting requirements, associated with successful drug candidates and decision-making throughout the drug discovery pipeline, including library design.

Abstract as supplied by the indexing source. © the authors / publisher.