Learning pharmacometric covariate model structures with symbolic regression networks
(2023) In Journal of Pharmacokinetics and Pharmacodynamics- Abstract
- Efficiently finding covariate model structures that minimize the need for random effects to describe pharmacological data is challenging. The standard approach focuses on identification of relevant covariates, and present methodology lacks tools for automatic identification of covariate model structures. Although neural networks could potentially be used to approximate covariate-parameter relationships, such approximations are not human-readable and come at the risk of poor generalizability due to high model complexity. In the present study, a novel methodology for simultaneous selection of covariate model structure and optimization of its parameters is proposed. It is based on symbolic regression, posed as an optimization problem with... (More)
- Efficiently finding covariate model structures that minimize the need for random effects to describe pharmacological data is challenging. The standard approach focuses on identification of relevant covariates, and present methodology lacks tools for automatic identification of covariate model structures. Although neural networks could potentially be used to approximate covariate-parameter relationships, such approximations are not human-readable and come at the risk of poor generalizability due to high model complexity. In the present study, a novel methodology for simultaneous selection of covariate model structure and optimization of its parameters is proposed. It is based on symbolic regression, posed as an optimization problem with smooth loss function. This enables training the model through back-propagation using efficient gradient computations. Feasibility and effectiveness is demonstrated by application to a clinical pharmacokinetic data set for propofol, containing infusion and blood sample time series from 1,031 individuals. The resulting model is compared to a published state-of-the-art model for the same data set. Our methodology finds a covariate model structure and corresponding parameter values with a slightly better fit, while relying on notably fewer covariates than the state-of the-art model. Unlike contemporary practice, finding the covariate model structure is achieved without an iterative procedure involving manual interactions. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/c153366c-3781-45f5-bb2b-411853e32320
- author
- Wahlquist, Ylva LU ; Sundell, Jesper LU and Soltesz, Kristian LU
- organization
- publishing date
- 2023
- type
- Contribution to journal
- publication status
- published
- subject
- in
- Journal of Pharmacokinetics and Pharmacodynamics
- publisher
- Springer
- external identifiers
-
- scopus:85174590660
- pmid:37864654
- ISSN
- 1567-567X
- DOI
- 10.1007/s10928-023-09887-3
- project
- Learning pharmacometric model structures from data
- language
- English
- LU publication?
- yes
- id
- c153366c-3781-45f5-bb2b-411853e32320
- alternative location
- https://rdcu.be/dpaAv
- date added to LUP
- 2023-06-21 09:33:59
- date last changed
- 2023-12-21 03:00:39
@article{c153366c-3781-45f5-bb2b-411853e32320, abstract = {{Efficiently finding covariate model structures that minimize the need for random effects to describe pharmacological data is challenging. The standard approach focuses on identification of relevant covariates, and present methodology lacks tools for automatic identification of covariate model structures. Although neural networks could potentially be used to approximate covariate-parameter relationships, such approximations are not human-readable and come at the risk of poor generalizability due to high model complexity. In the present study, a novel methodology for simultaneous selection of covariate model structure and optimization of its parameters is proposed. It is based on symbolic regression, posed as an optimization problem with smooth loss function. This enables training the model through back-propagation using efficient gradient computations. Feasibility and effectiveness is demonstrated by application to a clinical pharmacokinetic data set for propofol, containing infusion and blood sample time series from 1,031 individuals. The resulting model is compared to a published state-of-the-art model for the same data set. Our methodology finds a covariate model structure and corresponding parameter values with a slightly better fit, while relying on notably fewer covariates than the state-of the-art model. Unlike contemporary practice, finding the covariate model structure is achieved without an iterative procedure involving manual interactions.}}, author = {{Wahlquist, Ylva and Sundell, Jesper and Soltesz, Kristian}}, issn = {{1567-567X}}, language = {{eng}}, publisher = {{Springer}}, series = {{Journal of Pharmacokinetics and Pharmacodynamics}}, title = {{Learning pharmacometric covariate model structures with symbolic regression networks}}, url = {{http://dx.doi.org/10.1007/s10928-023-09887-3}}, doi = {{10.1007/s10928-023-09887-3}}, year = {{2023}}, }