Gordon Research Conference on Drug Metabolism 2019 | Holderness, NH
Date: July 7-12, 2019
Presenter: Neha Murad, PhD
Title: An Evaluation of In-silico Methods for Predicting Volume of Distribution in Humans for Greater than 300 Compounds
Abstract: Physiologically Based Pharmacokinetic (PBPK) modeling investigates the ADME (Absorption, Distribution, Metabolism, and Excretion) properties of a drug over a course of time. Volume of distribution (VD) is an important pharmacokinetic (PK) parameter that gives us information about the drug’s distribution in the body, it is also used to determine half-life life (T_(1/2)). These are crucial parameters to determine optimal drug dosage. There are several in silico mechanistic models that can be used to predict VD. Tissue-to-plasma partition coefficients (Kp) are another essential parameter needed to predict VD and thus in turn describe drug disposition. In this study, we explore four different modeling approaches to VD prediction. (i) Mechanistic Kp prediction models from in vitro properties, such as logP, pKa, fu, p and blood: plasma ratio. (ii) Tissue Level Kp prediction using Machine Learning. (iii) VD predictions using Machine Learning (iv) Human VD prediction using Allometric scaling. The use of any of these in silico methods in a priori human PK prediction for novel compounds requires a choice among these Mechanistic/ML PK models based on predictive value as well as the availability of predictive models for the underlying in vitro parameters (when needed) based on compound structure. While analyses of the predictive value of these in silico models have been performed, they typically rely on small compound sets (70-150) with sparse data, lack in-depth comparative analysis between models, and are rarely presented in a paradigm that can be utilized for prediction of de novo compounds. In this study we perform a thorough evaluation of the different in silico VDSS prediction methods using more than 300 structurally diverse compounds and present results which allude which are the more robust modeling approaches in predictive PK.