ACS Fall National Meeting & Exposition

This study presents novel QSAR models for the prediction of two key pharmacokinetic properties of drugs— the extent of plasma protein binding (PPB) and apparent volume of distribution (Vd) in humans. Experimental PPB data were represented by almost 1500 percentage bound values and about 300 human serum albumin affinity constants. Predictive models were developed using recently introduced GALAS modeling methodology that allows estimating reliability of resulting predictions and provides the basis for model trainability. Vd was modeled using a mechanistic approach accounting for drug binding in both plasma and tissues. 800 original Vd values collected from literature were corrected for free fraction in plasma yielding ‘unbound Vd’ (Vdu) values that represent drugs’ affinity to tissues. pVdu was then described by a nonlinear model in terms of simple physicochemical properties (logP and pK_a). Validation results indicate good predictive power of the obtained model with RMSE of pVdu prediction being 0.4 log units.

In recent years the pharma/biotech industries have made a concerted effort to focus on the quality of compounds produced in discovery versus quantity. Scientists are consequently required to consider all relevant properties of compounds to produce candidates with drug-like profiles. Balancing these inter-related molecular properties (such as solubility and permeability) can be a significant challenge. Tools that can help address these concerns are of considerable value. We will discuss case studies that illustrate how inter-related physical properties such as pK_a, logP, and logD can be modified to improve physical properties, such as solubility, using a software tool that combines physicochemical property predictors with a database of substituents. Structural modifications are addressed through subtle alterations, such as heterocyclic group replacement, functional group interchange, or more drastic changes, such as the addition of substituents.

This study presents a series of in-silico models for the prediction of probable inhibitors of CYP450 isoforms 3A4, 2D6, 2C9, 2C19, and 1A2 developed using a novel GALAS modeling methodology allowing estimation of prediction Reliability Indices. Inhibition constant thresholds of 10 and 50 µM were used to classify compounds in the initial data sets ranging from ca. 5000 to 8000 compounds for five considered enzyme isoforms. Obtained RI values correlate with prediction accuracy. Predictions with low RI are outside model applicability domain and cannot be considered. For the predictions with acceptable RI values, the accuracy approaches 90% in all five internal test sets. All models have been externally validated using the latest data from PubChem screening program. GALAS modeling methodology utilized in this work enables fast and efficient training of the obtained models, i.e., extending their applicability domain, adjusting them to screen proprietary databases for potential CYP inhibitors.