Skip to Main content Skip to Navigation
Theses

On learning mechanistic models from time series data with applications to personalised chronotherapies

Abstract : Mathematical modeling of biological processes aims at providing formal repre-sentations of complex systems to enable their study, both in a qualitative and quan-titative fashion. The need for explainability suggests the recourse to mechanisticmodels, which explicitly describe molecular interactions. Nevertheless, such mod-els currently rely on the existence of prior knowledge on the underlying reactionnetwork structure. Moreover, their conception remains an art which necessitatescreativity combined to multiple interactions with analysis and data fitting tools.This rules out numerous applications conceivable in personalized medicine, andcalls for methodological advances towards machine learning of patient-tailoredmodels. This thesis intends to devise algorithms to learn models of dynamicalinteractions from temporal data, with an emphasis on explainability for the humanmodeler. Its applications are in the context of personalized chronotherapies, thatconsist in optimizing drug administration with respect to the patient’s biologicalrhythms over the 24-hour span. Three main themes are explored: mechanisticmodeling, network inference and treatment personalization. The first chapter de-scribes the development of the first quantitative mechanistic model of the cellularcircadian clock integrating transcriptomic, proteomic and sub-cellular localizationdata. This model has been successfully connected to a model of cellular pharmacol-ogy of an anticancerous drug, irinotecan, achieving personalization of its optimaladministration timing. The second chapter introduces a novel protocol for inferringwhole-body systemic controls enforced on peripheral clocks. On the long run, thisapproach will make it possible to integrate individual data collected from wearablesfor personalized chronotherapies. The third chapter presents a general algorithmto infer reactions with chemical kinetics from time series data.
Complete list of metadata

https://tel.archives-ouvertes.fr/tel-03686289
Contributor : ABES STAR :  Contact
Submitted on : Thursday, June 2, 2022 - 3:47:13 PM
Last modification on : Monday, September 5, 2022 - 3:57:22 PM
Long-term archiving on: : Saturday, September 3, 2022 - 7:47:35 PM

File

103858_MARTINELLI_2022_archiva...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03686289, version 1

Citation

Julien Martinelli. On learning mechanistic models from time series data with applications to personalised chronotherapies. Machine Learning [stat.ML]. Institut Polytechnique de Paris, 2022. English. ⟨NNT : 2022IPPAX011⟩. ⟨tel-03686289⟩

Share

Metrics

Record views

116

Files downloads

67