Prediction of Compound Plasma Concentration–Time Profiles in Mice Using Random Forest

Abstract

Pharmacokinetic (PK) parameters such as clearance (CL) and volume of distribution (Vd) have been the subject of previous in silico predictive models. However, having information of the concentration over time profile explicitly can provide additional value like time above MIC or AUC, etc., to understand both the efficacy and safety-related aspects of a compound. In this work, we developed machine learning models for plasma concentration–time profiles after both i.v. and p.o. dosing for a series of 17 in-house projects. For explanatory variables, MACCS Keys chemical descriptors as well as in silico and experimental in vitro PK parameters were used. The predictive accuracy of random forest (RF), message passing neural network, 2-compartment models using estimated CL and Vdss, and an average model (as a control experiment) was investigated using 5-fold cross-validation (5-fold CV) and leave-one-project-out validation (LOPO-V). The predictive accuracy of RF in 5-fold CV for i.v. and p.o. plasma concentration–time profiles was the best among the models studied, with an RMSE for i.v. dosing at 0.08, 1, and 8 h of 0.245, 0.474, and 0.462, respectively, and an RMSE for p.o. dosing at 0.25, 1, and 8 h of 0.500, 0.612, and 0.509, respectively. Furthermore, by investigating the importance of the in vitro PK parameters using the Gini index, we observed that the general prior knowledge in ADME research was reflected well in the respective feature importance of in vitro parameters such as predicted human Vd (hVd) for the initial distribution, mouse intrinsic CL and unbound fraction of mouse plasma for the elimination process, and Caco2 permeability for the absorption process. Also, this model is the first model that can predict twin peaks in the concentration–time profile much better than a baseline compartment model. Because of its combination of sufficient accuracy and speed of prediction, we found the model to be fit-for-purpose for practical lead optimization.

Systematic Evaluation of High-Throughput PBK Modelling Strategies for the Prediction of Intravenous and Oral Pharmacokinetics in Humans

“…Potential applications of such HT-PBK modelling strategies are vast, and there have been many efforts recently in both pharmacology and toxicology to establish such strategies for different use cases and based on different approaches. Many of them, however, relied exclusively on rodent data for their validation (Schneckener et al 2019; Kamiya et al 2021; Naga et al 2022; Punt et al 2022b; Obrezanova et al 2022; Mavroudis et al 2023; Handa et al 2023; Führer et al 2024). Others did perform predictions for humans but relied in their validation of prediction quality on summary PK parameters, like Cmax or AUC values (Punt et al 2022a; Miljković et al 2021; Li et al 2023).…”

“…Other times, it is done by predicting mechanistically relevant compound properties, like the lipophilicity, solubility or clearance of compounds, which can then be used as inputs for making PK predictions using mechanistic models (Danishuddin et al 2022; Pillai et al 2022; Fagerholm et al 2023; Mavroudis et al 2023; Führer et al 2024). Until now, many in silico -based PK prediction efforts have focused on predicting rodent data (Schneckener et al 2019; Kamiya et al 2021; Naga et al 2022; Punt et al 2022b; Obrezanova et al 2022; Mavroudis et al 2023; Handa et al 2023), presumably due to its greater availability than human data. While valuable, the ultimate goal in pharmacology and toxicology is to yield human-relevant conclusions.…”

“…Until now, many in silico-based PK prediction efforts have focused on predicting rodent data (Schneckener et al 2019;Kamiya et al 2021;Naga et al 2022;Punt et al 2022b;Obrezanova et al 2022;Mavroudis et al 2023;Handa et al 2023), presumably due to its greater availability than human data. While valuable, the ultimate goal in pharmacology and toxicology is to yield human-relevant conclusions.…”

Prediction of Inhibitory Activity Against the MATE1 Transporter via Combined Fingerprint- and Physics-Based Machine Learning Models

“…We next investigated the applicability domain of two RF models, based on only ECFP4 fingerprints, and based on both ECFP4 and MM-GB/SA scores,by 5-NN analysis using Tanimoto similarity of ECFP4 fingerprints. According to one practical implementation, if more compounds are predicted correctly in a higher Tanimoto similarity bin in a 5-NN analysis, we can say this model has an interpretable applicability domain, which we can use to estimate prediction confidence for a new compound in a given 5NN similarity bin 51,52 . The results of this analysis are shown in Figure 4A where it can be seen that in RF model using only ECFP4, the percentage of correctly predicted compounds increases according to the Tanimoto similarity, from 71% in the 0.1 to 0.2 of Tanimoto similarity bin (TSB), over 73% (0.2 to 0.3 of TSB) and 84% (0.3 to 0.4 of TSB), to 85% (more than 0.4 of TSB).…”

Combined data-driven and mechanism-based approaches for human-intestinal-absorption prediction in the early drug-discovery stage

“…is more difficult than that of intravenous (i.v.). 24 Recently, two studies were conducted to predict Fa. In the first study in 2019, the authors aimed to use multiple-class classifiers using Caco-2 permeability and DMSO solubility as explanatory variables, and finally they built a 3-class classifier with high predictability with accuracy and kappa values of 0.836 and 0.560, respectively.…”