In silico prediction of log P for small drug molecules: quo vadis?

Abstract

Lipophilicity or hydrophobicity, expressed as the n-octanol/water partition coefficient (log P), is a fundamental physicochemical property in drug discovery. Although experimental determination methods such as shake-flask, slow-stirring, and RP-HPLC remain the gold standard, they are often labor-intensive, costly, and limited in throughput. Consequently, in silico prediction of log P has become indispensable, particularly for virtual screening of vast chemical libraries. Over the past decades, diverse computational strategies, including atom-based, fragment-based, topology-based, and structural property-based methods, have been developed, with recent advances driven by artificial intelligence and machine learning (AI/ML). Despite these efforts, no single model consistently outperforms others across chemical classes, and challenges such as data quality, applicability domain limitations, conformational diversity, and systematic error persist. This review critically examines conventional and modern approaches to log P prediction, evaluates their strengths and limitations, and discusses emerging trends aimed at improving predictive accuracy and interpretability in drug discovery. KEYWORDS artificial intelligence (AI), machine learning (ML), in silico, lipophilicity, log P.

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