Computational Predictions of Nonclinical Pharmacokinetics at the Drug Design Stage

Abstract

Although computational predictions of pharmacokinetics (PK) are desirable at the drug design stage, existing approaches are often limited by prediction accuracy and human interpretability. Using a discovery data set of mouse and rat PK studies at Roche (9,685 unique compounds), we performed a proof-of-concept study to predict key PK properties from chemical structure alone, including plasma clearance (CLp), volume of distribution at steady-state (Vss), and oral bioavailability (F). Ten machine learning (ML) models were evaluated, including Single-Task, Multitask, and transfer learning approaches (i.e., pretraining with in vitro data). In addition to prediction accuracy, we emphasized human interpretability of outcomes, especially the quantification of uncertainty, applicability domains, and explanations of predictions in terms of molecular features. Results show that intravenous (IV) PK properties (CLp and Vss) can be predicted with good precision (average absolute fold error, AAFE of 1.96− 2.84 depending on data split) and low bias (average fold error, AFE of 0.98−1.36), with AutoGluon, Gaussian Process Regressor (GP), and ChemProp displaying the best performance. Driven by higher complexity of oral PK studies, predictions of F were more challenging, with the best AAFE values of 2.35−2.60 and higher overprediction bias (AFE of 1.45−1.62). Multi-Task approaches and pretraining of ChemProp neural networks with in vitro data showed similar precision to Single-Task models but helped reduce the bias and increase correlations between observations and predictions. A combination of GP-computed prediction variance, molecular clustering, and dimensionality-reduction provided valuable quantitative insights into prediction uncertainty and applicability domains. SHAPley Additive exPlanations (SHAPs) highlighted molecular features contributing to prediction outcomes of Vss, providing explanations that could aid drug design. Combined results show that computational predictions of PK are feasible at the drug design stage, with several ML technologies converging to successfully leverage historical PK data sets. Further studies are needed to unlock the full potential of this approach, especially with respect to data set sizes and quality, transfer learning between in vitro and in vivo data sets, model-independent quantification of uncertainty, and explainability of predictions.

A Machine Learning Framework to Improve Rat Clearance Predictions and Inform Physiologically Based Pharmacokinetic Modeling

“…However, it is challenging on small datasets because the algorithm has to learn a molecular embedding from scratch. With the help of transfer learning, embeddings that had been optimized on larger datasets, e.g., from a CL int,in vitro,app model, could be reused and only slightly be refined during the training of our ML models . Such investigations remain out of scope for this work.…”

“…With the help of transfer learning, embeddings that had been optimized on larger datasets, e.g., from a CL int,in vitro,app model, could be reused and only slightly be refined during the training of our ML models. 13 Such investigations remain out of scope for this work. Nevertheless, our study shows the benefit of using ML models in combination with WSM/PBPK modeling as existing workflows can be reused and improved.…”

“…To minimize the bias effects in IVIVE as well as for application in early stages of development before in vitro data is generated, ML models for clearance prediction can be valuable. Some approaches for ML models for in vitro and in vivo clearance prediction are available in the literature. ,− For example, Kosugi et al built random forest and radial basis function ML models to predict the total plasma clearance in rats and compared the ML predictions to bottom-up IVIVE from microsomal CL int using the WSM. The bottom-up approach tended to underestimate the total plasma clearance, which could be improved with the ML approach although it has to be noted that alternative scaling methods were employed to improve the bottom-up estimation.…”

“…For a larger collection of computational studies that predicted nonclinical and clinical PK properties, we refer to Stoyanova et al…”

“…For a larger collection of computational studies that predicted nonclinical and clinical PK properties, we refer to Stoyanova et al 13 Even if ML models for PK prediction in humans perform well on clinical data, the training data is limited 16 and will likely not represent the preclinical space well. This is because PK is often optimized preclinically to obtain values in a desired range, so data for molecules with unfavorable PK parameters are missing in clinical ML training sets.…”