Development and Implementation of an enterprise-wide Predictive Model for Early absorption, distribution, Metabolism and Excretion Properties

Abstract

Background: Accurate prediction of absorption, distribution, metabolism and excretion (ADME) properties can facilitate the identification of promising drug candidates. Methodology & Results: The authors present the Janssen generic Target Product Profile (gTPP) model, which predicts 18 early ADME properties, employs a graph convolutional neural network algorithm and was trained on between 1000–10,000 internal data points per predicted parameter. gTPP demonstrated stronger predictive power than pretrained commercial ADME models and automatic model builders. Through a novel logging method, the authors report gTPP usage for more than 200 Janssen drug discovery scientists. Conclusion: The investigators successfully enabled the rapid and systematic implementation of predictive ML tools across a drug discovery pipeline in all therapeutic areas. This experience provides useful guidance for other large-scale AI/ML deployment efforts.

Predictive Models Based on Molecular Images and Molecular Descriptors for Drug Screening

“…To reduce the use of animal experiments and developmental expenses, predictive models based on quantitative structure–activity relationship (QSAR) analysis have been actively used in recent years at the drug discovery stage. − QSAR analysis is an approach based on features of compounds such as molecular descriptors (MDs) and fingerprints, and algorithms such as support vector machine, random forest, artificial neural network, k-nearest neighbor, XGBoost, and deep learning (DL) have been applied. − At the drug discovery stage, QSAR analysis is used to predict pharmacological activity, pharmacokinetic parameters, and toxicity parameters. Several predictive models based on QSAR analysis were previously reported for parameters associated with toxicity, such as inhibition of the human ether-a-go-go related gene and 50% lethal dose (LD 50 ), and parameters associated with pharmacokinetics, such as metabolic stability, protein binding, distribution to blood cells, membrane permeability, clearance (CL), and the CL pathway. − Therefore, these predictive models are used by various pharmaceutical companies to perform virtual screening, narrow hit compounds, and prioritize various experiments. − , Thus, predictive models have been reported for many toxicity parameters and pharmacokinetic parameters. − , However, for some targets, the predictive performance of these models is apparently insufficient. As an approach to improve the predictive performance for targets such as LD 50 , a consensus model that combines predictive models constructed by 35 different organizations was reported .…”

“…[1][2][3]12 Thus, predictive models have been reported for many toxicity parameters and pharmacokinetic parameters. [1][2][3]12 However, for some targets, the predictive performance of these models is apparently insufficient. As an approach to improve the predictive performance for targets such as LD 50 , a consensus model that combines predictive models constructed by 35 different organizations was reported.…”

“…To reduce the use of animal experiments and developmental expenses, predictive models based on quantitative structure–activity relationship (QSAR) analysis have been actively used in recent years at the drug discovery stage. 1 − 3 QSAR analysis is an approach based on features of compounds such as molecular descriptors (MDs) and fingerprints, and algorithms such as support vector machine, random forest, artificial neural network, k-nearest neighbor, XGBoost, and deep learning (DL) have been applied. 1 − 4 At the drug discovery stage, QSAR analysis is used to predict pharmacological activity, pharmacokinetic parameters, and toxicity parameters.…”

“… 5 − 11 Therefore, these predictive models are used by various pharmaceutical companies to perform virtual screening, narrow hit compounds, and prioritize various experiments. 1 − 3 , 12 Thus, predictive models have been reported for many toxicity parameters and pharmacokinetic parameters. 1 − 3 , 12 However, for some targets, the predictive performance of these models is apparently insufficient.…”

Enabling Deoxygenative C(sp2)-C(sp3) Cross-Coupling for Parallel Medicinal Chemistry

“…In PMC, the diversity of synthesized compounds is directly linked to the chemical space of building blocks being utilized. − Therefore, in considering a synthetic method for library synthesis, the availability and chemical diversity of its reagent pool becomes imperative. − We conducted cheminformatic analysis on common alkyl precursors for C­(sp 2 )-C­(sp 3 ) coupling reported in literature including boronates, trifluoroborate salts, alkyl halides (from which organozinc reagents may be derived), carboxylic acids and alcohols (Figure A). , Specifically, t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction of the first 50 principal components using 512-bit Morgan fingerprint descriptors was employed to quantify structural similarities and visualize the commercially accessible chemical space for each sp 3 building block. − As shown in Figure A, alkyl boronates, including potassium trifluoroborate salts, are commercially available in limited numbers and represent limited chemical diversity. A significant number of alkyl halide building blocks are commercially available, but there are significant gaps in their chemical space coverage.…”