Rs-predictor --- creation of cytochrome p450 regioselectivity models
Curt M. Breneman, Jed Mikhail Zarelzki · 2011
This thesis describes RS-Predictor, a new in silico method for generating predictive models of P450-mediated metabolism for drug-like compounds. Within this method, potential sites of metabolism (SOMs) are represented as metabolophores: A concept that describes the heirarchical combination of topological and quantum chemical descriptors needed to represent the reactivity of potential metabolic reaction sites. RS-Predictor modeling involves the use of metabolophore descriptors together with multiple-instance ranking (MIRank) to generate an optimized descriptor weight vector that encodes regioselectivity trends across all cases in a training set. The resulting pathway-independent (ex. O-dealkylation versus Csp3 Hydroxylation), isozyme-specific regioselectivity model to may be used to predict potential metabolic liabilities. In one of the first applications of rank aggregation within the chemoinformatics community, independently-generated regioselectivity rankings for a given compound are merged into single optimized consensus predictions. A new Lift metric for assessing prediction quality is introduced, where each substrate is assigned a lift weight that expresses the statistical likelihood of randomly picking the CYP-oxidized SOM(s) out of all putative SOMs on the substrate. The prediction quality of each model is also assessed through the number of correct and incorrect predictions made on a pathway-by-pathway basis. The broad applicability of RS-Predictor is demonstrated through the creation of regioselectivity models for substrate sets of the following CYPs: 1A2(271), 2A6(105), 2B6(151), 2C19(218), 2C8(142), 2C9(226), 2D6(270), 2E1(145) and 3A4(475), as well as a Merged set of all 680 curated substrates. A comprehensive investigation into the relative signal content of descriptors from different classes for each isozyme is made through the generation of seven separate RS-Predictor models for each substrate set. Two of these models involve the incorporation of high quality DFT derived reactivity information from SMARTCyp, a technology developed separately by another research group. Optimal combinations of RS-Predictor and SMARTCyp are shown to have stronger performances than either method alone, correctly identifying a large proportion of the metabolites of each dataset in the top two rank-positions: 1A2(83.0%), 2A6(85.7%), 2B6(82.1%), 2C19(86.2%), 2C8(83.8%), 2C9(84.5%), 2D6(85.9%), 2E1(82.8%), 3A4(82.3%), Merged(86.0%). A cross-isozyme (CI) study is made through the application of the regioselectivity QSAR of one isozyme to predict the CYP-mediated metabolism of the substrates of a different isozyme. Comparing CI model performances against the original results of cross-validated (CV) models for the same set of substrates lets future users of RS-Predictor gauge the relative benefits of creating isozyme-specific models.