An information-theoretic approach to integrated mechanistic-empirical modeling of cellular response based on intracellular signaling dynamics

Abstract

The following paper presents a new systematic approach to the design and construction of a hybrid mechanistic-empirical model for the prediction of cellular response to extracellular cues. The hybrid framework incorporates computable biological models, such as signal transduction network, with empirical experimental data. The environment input cues are augmented by intracellular signals computed as simulated response to input cues. The mechanistic model of signal transduction, however, is often too complex to predict downstream cell behaviors, or the details of the downstream signaling events are not accurately known. To fill the gap we incorporate an empirical model that relates the augmented input space of extracellular cues(computed using the mechanistic model) to an observable output space using Partial Least Squares Regression (PLSR). Akaikie's Information Criterion is used to find an optimal order of the PLSR model based on the trade-off between accuracy and variance. This two-stage approach (first augmenting the input space through a mechanistic map, then eliminating co-linearity and empirically correlating to downstream behaviors by PLSR) is a powerful tool for this class of integrated mechanistic-empirical modeling problems. We first introduce the framework of the mechanistic-empirical hybrid model, present an AIC-based model structure metric, and apply the method to a T-Cell immuno-response problem. The resultant lower-order, nonlinear, mechanistic-empirical model that accurately represents the process being studied.