A Comparative Study of Arbitrary Versus Tailored Molecular Similarity Metrics in Property / Toxicity / Bioactivity Prediction

摘要

Quantitative molecular similarity analysis (QMSA) methods using a variety of calculated molecular descriptors and experimental properties have been used to select analogs and estimate a wide variety of properties pertaining to chemistry, pharmacology, and toxicology. Traditionally, descriptor sets for QMSA methods have been selected intuitively by an expert. We developed 'tailoring' as a new approach to QMSA that selects and uses indices strongly correlated with the property of interest, thus creating a property-specific molecular similarity space. This presentation will discuss various tailoring methods used to select indices and the improvement in property estimation resulting from these techniques.