Alexander Janssen, Frank C Bennis, Marjon H Cnossen, Ron A A Mathôt
{"title":"On inductive biases for the robust and interpretable prediction of drug concentrations using deep compartment models.","authors":"Alexander Janssen, Frank C Bennis, Marjon H Cnossen, Ron A A Mathôt","doi":"10.1007/s10928-024-09906-x","DOIUrl":null,"url":null,"abstract":"<p><p>Conventional pharmacokinetic (PK) models contain several useful inductive biases guiding model convergence to more realistic predictions of drug concentrations. Implementing similar biases in standard neural networks can be challenging, but might be fundamental for model robustness and predictive performance. In this study, we build on the deep compartment model (DCM) architecture by introducing constraints that guide the model to explore more physiologically realistic solutions. Using a simulation study, we show that constraints improve robustness in sparse data settings. Additionally, predicted concentration-time curves took on more realistic shapes compared to unconstrained models. Next, we propose the use of multi-branch networks, where each covariate can be connected to specific PK parameters, to reduce the propensity of models to learn spurious effects. Another benefit of this architecture is that covariate effects are isolated, enabling model interpretability through the visualization of learned functions. We show that all models were sensitive to learning false effects when trained in the presence of unimportant covariates, indicating the importance of selecting an appropriate set of covariates to link to the PK parameters. Finally, we compared the predictive performance of the constrained models to previous relevant population PK models on a real-world data set of 69 haemophilia A patients. Here, constrained models obtained higher accuracy compared to the standard DCM, with the multi-branch network outperforming previous PK models. We conclude that physiological-based constraints can improve model robustness. We describe an interpretable architecture which aids model trust, which will be key for the adoption of machine learning-based models in clinical practice.</p>","PeriodicalId":16851,"journal":{"name":"Journal of Pharmacokinetics and Pharmacodynamics","volume":" ","pages":"355-366"},"PeriodicalIF":2.2000,"publicationDate":"2024-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11255087/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Pharmacokinetics and Pharmacodynamics","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1007/s10928-024-09906-x","RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2024/3/26 0:00:00","PubModel":"Epub","JCR":"Q3","JCRName":"PHARMACOLOGY & PHARMACY","Score":null,"Total":0}
引用次数: 0
Abstract
Conventional pharmacokinetic (PK) models contain several useful inductive biases guiding model convergence to more realistic predictions of drug concentrations. Implementing similar biases in standard neural networks can be challenging, but might be fundamental for model robustness and predictive performance. In this study, we build on the deep compartment model (DCM) architecture by introducing constraints that guide the model to explore more physiologically realistic solutions. Using a simulation study, we show that constraints improve robustness in sparse data settings. Additionally, predicted concentration-time curves took on more realistic shapes compared to unconstrained models. Next, we propose the use of multi-branch networks, where each covariate can be connected to specific PK parameters, to reduce the propensity of models to learn spurious effects. Another benefit of this architecture is that covariate effects are isolated, enabling model interpretability through the visualization of learned functions. We show that all models were sensitive to learning false effects when trained in the presence of unimportant covariates, indicating the importance of selecting an appropriate set of covariates to link to the PK parameters. Finally, we compared the predictive performance of the constrained models to previous relevant population PK models on a real-world data set of 69 haemophilia A patients. Here, constrained models obtained higher accuracy compared to the standard DCM, with the multi-branch network outperforming previous PK models. We conclude that physiological-based constraints can improve model robustness. We describe an interpretable architecture which aids model trust, which will be key for the adoption of machine learning-based models in clinical practice.
期刊介绍:
Broadly speaking, the Journal of Pharmacokinetics and Pharmacodynamics covers the area of pharmacometrics. The journal is devoted to illustrating the importance of pharmacokinetics, pharmacodynamics, and pharmacometrics in drug development, clinical care, and the understanding of drug action. The journal publishes on a variety of topics related to pharmacometrics, including, but not limited to, clinical, experimental, and theoretical papers examining the kinetics of drug disposition and effects of drug action in humans, animals, in vitro, or in silico; modeling and simulation methodology, including optimal design; precision medicine; systems pharmacology; and mathematical pharmacology (including computational biology, bioengineering, and biophysics related to pharmacology, pharmacokinetics, orpharmacodynamics). Clinical papers that include population pharmacokinetic-pharmacodynamic relationships are welcome. The journal actively invites and promotes up-and-coming areas of pharmacometric research, such as real-world evidence, quality of life analyses, and artificial intelligence. The Journal of Pharmacokinetics and Pharmacodynamics is an official journal of the International Society of Pharmacometrics.