Optimization for Explainable Modeling

Loading...
Thumbnail Image

Date

2023-05

Authors

Journal Title

Journal ISSN

Volume Title

Publisher

The Ohio State University

Research Projects

Organizational Units

Journal Issue

Abstract

Whether it is the signaling mechanisms behind immune cells or the change in animal populations, mechanistic models such as agent based models or systems of differential equations that define explicit causal mechanisms are used to validate hypothesises and thereby understand physical systems. To quantitatively and qualitatively validate a mechanistic model, experimental data is used to fit and estimate parameters within these models, thereby providing interpretable and explainable quantitative values. Parameter estimation tasks for mechanistic models can be extremely challenging for a variety of reasons, especially for single-cell systems. One, measurements of protein abundances can vary many orders of magnitude and often the number of model parameters exceeds that of the data. Two, mechanistic simulations can often be computationally expensive where parameter estimation can range from hours to days, and even more when fine-tuning an optimization algorithm. Through building a framework BioNetGMMFit, we show that we can readily account for the large variances within single-cell models using generalized method of moments, and through leveraging deep learning in surrogate modeling, we show that we can reduce the computational time complexity in parameter estimation.

Description

Keywords

Citation