Identification of cell‐to‐cell heterogeneity through systems engineering approaches

Abstract

Cells in a genetically homogeneous cell‐population exhibit a significant degree of heterogeneity in their responses to an external stimulus. To understand origins and importance of this heterogeneity, individual‐based population model (IBPM), where parameters follow probability density functions (PDFs) instead of being constants, has been previously developed. However, parameter identification for an IBPM is challenging as estimating PDFs is computationally expensive. Also, because of experimental limitations and nonlinearity of models, not all parameters' PDFs are identifiable. Motivated by the above considerations, a new methodology is proposed in this study. First, a subset of parameters whose PDFs is identifiable are determined through sensitivity analysis, and only these PDFs are estimated. Second, an artificial neural network model is developed to find an empirical relation between these parameter and output PDFs to reduce computational costs of the parameter identification. The proposed approach is validated by estimating PDFs of parameters of a tumor necrosis factor‐α signaling model.

Development of a hybrid model for a partially known intracellular signaling pathway through correction term estimation and neural network modeling

“…where w s (t) is the continuous temporal profile of w(t) from t = 0 to t N t under input u s , and y i ðu s ; tÞ is the i th output measured under input u s at time t. However, Eq 3 is likely to be ill-conditioned because (1) it is an infinite dimensional problem in which the decision variables (i.e., w(t)) are continuous temporal profiles, and (2) the available measurements are limited in quantity (i.e., small values of n y and N t ) [33][34][35][36]. Hence, its solution is subject to high uncertainties, and the resulted hybrid model based on the estimated w(t) will be difficult to be generalized for future predictions.…”

HybridPDE‐kMCmodeling approach to simulate multivalent lectin‐glycan binding process

“…It is known that a stochastic model is generally computationally more demanding compared to its deterministic counterpart. 36,42,43 This issue is usually exacerbated for a spatial stochastic model such as the spatial kMC model discussed earlier. The main reason for the increased computational cost is that the proposed kMC model explicitly considers the temporal evolution of the cell membrane, by simulating glycan diffusion on the surface, and uses this information for computing probabilities of surface reactions.…”

“…One remaining challenge associated with the kMC model proposed for studying lectin‐glycan binding processes is the computational efficiency. It is known that a stochastic model is generally computationally more demanding compared to its deterministic counterpart 36,42,43 . This issue is usually exacerbated for a spatial stochastic model such as the spatial kMC model discussed earlier.…”

Operable adaptive sparse identification of systems: Application to chemical processes

“…In the last two decades, DNNs have been widely used for various chemical engineering applications such as process control, 23,24 fault diagnosis, 25 system identification, 26,27 sensor data analysis, 28 and process design and simulation 29 . This success can be attributed to their ability to learn and approximate any underlying complex nonlinearities using a simple architecture.…”