A0842
Title: Validation of ODE models for biological networks through Bayesian approach
Authors: Donghui Son - Simon Fraser University (Canada)
Jaejik Kim - Sungkyunkwan University (Korea, South) [presenting]
Abstract: Ordinary differential equations (ODEs) are often used to describe the dynamics of complex networks or processes over time in various fields. They are typically derived from theoretical assumptions, known interactions, or patterns of observed data. However, since ODEs are inherently deterministic, they may not adequately capture the variability and noise present in data. Thus, the model fit might not capture all possible data variations, and there might be a discrepancy between model prediction and actual dynamics, potentially resulting in misleading interpretations. Moreover, it might be more serious, especially in biological contexts where multiple sources of uncertainty and low signal-to-noise ratio exist. Therefore, it is necessary to validate ODE models by considering such uncertainty and noise. Bayesian methods offer a powerful framework for capturing and quantifying these uncertainties. A Bayesian validation approach that explicitly estimates the model inadequacies of ODEs as bias functions of time is proposed. By estimating the bias functions across the entire observed time interval, the proposed method can quantify bias for all subintervals of time. This enables a more accurate and interpretable assessment of the model's performance and allows for improved predictive capabilities by systematically correcting for model bias.
