Speaker
Description
Objectives: Mechanistic models describe biological processes over time, but they often rely on few observed compartments and sparse longitudinal data. They may therefore be too simple to capture complex processes, such as post-vaccination immune dynamics, or may suffer from identifiability issues, especially in nonlinear mixed-effects models based on differential equations. At the same time, longitudinal high-throughput data, including transcriptomics and proteomics, are increasingly available and may help inform unobserved biological processes. Integrating such high-dimensional data into mechanistic models with latent compartments remains difficult.
Methods: We hypothesize that observed omics biomarkers can inform the dynamics of unobserved immune compartments. We propose a regularized estimation method for mechanistic models with latent compartments measured indirectly through high-dimensional longitudinal biomarkers. Relevant biomarkers are selected by regularizing the parameters linking them to latent compartments while estimating population mechanistic parameters. The algorithm alternates between a regularization step, based on penalized log-likelihood derivatives, and a mechanistic inference step using SAEM in Monolix.
Results: We evaluated the method in simulations and applied it to immune responses after Pfizer/BioNTech COVID-19 vaccination in 15 infection-naïve individuals (Rinchai et al.). Daily blood samples were collected for 9 days after each dose, with 8,172 gene expression measurements grouped into 34 pathways, and serology at baseline, day 7, and day 14. Inflammation, neutrophils, interferon, and type I interferon were most strongly associated with short-term immune response dynamics.
Conclusion: Transcriptomic data improved identifiability of latent compartments. Limitations include the assumed linear biomarker-compartment relationship and high computational cost. The method is implemented in the REMixed R package on CRAN.
References:
@article{tibshirani1996regression,
title={Regression shrinkage and selection via the lasso},
author={Tibshirani, Robert},
journal={Journal of the Royal Statistical Society Series B: Statistical Methodology},
volume={58},
number={1},
pages={267--288},
year={1996},
publisher={Oxford University Press}
}
@article{kuhn2005maximum,
title={Maximum likelihood estimation in nonlinear mixed effects models},
author={Kuhn, Estelle and Lavielle, Marc},
journal={Computational statistics \& data analysis},
volume={49},
number={4},
pages={1020--1038},
year={2005},
publisher={Elsevier}
}
@article{rinchai2022high,
title={High--temporal resolution profiling reveals distinct immune trajectories following the first and second doses of COVID-19 mRNA vaccines},
author={Rinchai, Darawan and Deola, Sara and Zoppoli, Gabriele and Kabeer, Basirudeen Syed Ahamed and Taleb, Sara and Pavlovski, Igor and Maacha, Selma and Gentilcore, Giusy and Toufiq, Mohammed and Mathew, Lisa and others},
journal={Science advances},
volume={8},
number={45},
pages={eabp9961},
year={2022},
publisher={American Association for the Advancement of Science}
}
