Loss formulations for assumption-free neural inference of SDE coefficient functions

Abstract Stochastic differential equations (SDEs) are one of the most commonly studied probabilistic dynamical systems, and widely used to model complex biological processes. Building upon the previously introduced idea of performing inference of dynamical systems by parametrising their coefficient...

Full description

Saved in:
Bibliographic Details
Main Authors: Marc Vaisband, Valentin von Bornhaupt, Nina Schmid, Izdar Abulizi, Jan Hasenauer
Format: Article
Language:English
Published: Nature Portfolio 2025-03-01
Series:npj Systems Biology and Applications
Online Access:https://doi.org/10.1038/s41540-025-00500-6
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849767274375806976
author Marc Vaisband
Valentin von Bornhaupt
Nina Schmid
Izdar Abulizi
Jan Hasenauer
author_facet Marc Vaisband
Valentin von Bornhaupt
Nina Schmid
Izdar Abulizi
Jan Hasenauer
author_sort Marc Vaisband
collection DOAJ
description Abstract Stochastic differential equations (SDEs) are one of the most commonly studied probabilistic dynamical systems, and widely used to model complex biological processes. Building upon the previously introduced idea of performing inference of dynamical systems by parametrising their coefficient functions via neural networks, we propose a novel formulation for an optimisation objective that combines simulation-based penalties with pseudo-likelihoods. This greatly improves prediction performance compared to the state-of-the-art, and makes it possible to learn a wide variety of dynamics without any prior assumptions on analytical structure.
format Article
id doaj-art-49fe3ec723e94a8f8fbb0ec333b22b0d
institution DOAJ
issn 2056-7189
language English
publishDate 2025-03-01
publisher Nature Portfolio
record_format Article
series npj Systems Biology and Applications
spelling doaj-art-49fe3ec723e94a8f8fbb0ec333b22b0d2025-08-20T03:04:16ZengNature Portfolionpj Systems Biology and Applications2056-71892025-03-0111111010.1038/s41540-025-00500-6Loss formulations for assumption-free neural inference of SDE coefficient functionsMarc Vaisband0Valentin von Bornhaupt1Nina Schmid2Izdar Abulizi3Jan Hasenauer4Department of Internal Medicine III with Haematology, Medical Oncology, Haemostaseology, Infectiology and Rheumatology, Oncologic Center, Salzburg Cancer Research Institute - Laboratory for Immunological and Molecular Cancer Research (SCRI-LIMCR)Bonn Center for Mathematical Life Sciences, Life & Medical Sciences (LIMES) Institute, University of BonnBonn Center for Mathematical Life Sciences, Life & Medical Sciences (LIMES) Institute, University of BonnBonn Center for Mathematical Life Sciences, Life & Medical Sciences (LIMES) Institute, University of BonnBonn Center for Mathematical Life Sciences, Life & Medical Sciences (LIMES) Institute, University of BonnAbstract Stochastic differential equations (SDEs) are one of the most commonly studied probabilistic dynamical systems, and widely used to model complex biological processes. Building upon the previously introduced idea of performing inference of dynamical systems by parametrising their coefficient functions via neural networks, we propose a novel formulation for an optimisation objective that combines simulation-based penalties with pseudo-likelihoods. This greatly improves prediction performance compared to the state-of-the-art, and makes it possible to learn a wide variety of dynamics without any prior assumptions on analytical structure.https://doi.org/10.1038/s41540-025-00500-6
spellingShingle Marc Vaisband
Valentin von Bornhaupt
Nina Schmid
Izdar Abulizi
Jan Hasenauer
Loss formulations for assumption-free neural inference of SDE coefficient functions
npj Systems Biology and Applications
title Loss formulations for assumption-free neural inference of SDE coefficient functions
title_full Loss formulations for assumption-free neural inference of SDE coefficient functions
title_fullStr Loss formulations for assumption-free neural inference of SDE coefficient functions
title_full_unstemmed Loss formulations for assumption-free neural inference of SDE coefficient functions
title_short Loss formulations for assumption-free neural inference of SDE coefficient functions
title_sort loss formulations for assumption free neural inference of sde coefficient functions
url https://doi.org/10.1038/s41540-025-00500-6
work_keys_str_mv AT marcvaisband lossformulationsforassumptionfreeneuralinferenceofsdecoefficientfunctions
AT valentinvonbornhaupt lossformulationsforassumptionfreeneuralinferenceofsdecoefficientfunctions
AT ninaschmid lossformulationsforassumptionfreeneuralinferenceofsdecoefficientfunctions
AT izdarabulizi lossformulationsforassumptionfreeneuralinferenceofsdecoefficientfunctions
AT janhasenauer lossformulationsforassumptionfreeneuralinferenceofsdecoefficientfunctions