Calibrating agent-based models to tumor images using representation learning

Agent-based models (ABMs) have enabled great advances in the study of tumor development and therapeutic response, allowing researchers to explore the spatiotemporal evolution of the tumor and its microenvironment. However, these models face serious drawbacks in the realm of parameterization – ABM parameters are typically set individually based on various data and literature sources, rather than through a rigorous parameter estimation approach. While ABMs can be fit to simple time-course data (such as tumor volume), that type of data loses the spatial information that is a defining feature of ABMs. While tumor images provide spatial information, it is exceedingly difficult to compare tumor images to ABM simulations beyond a qualitative visual comparison. Without a quantitative method of comparing the similarity of tumor images to ABM simulations, a rigorous parameter fitting is not possible. Here, we present a novel approach that applies neural networks to represent both tumor images and ABM simulations as low dimensional points, with the distance between points acting as a quantitative measure of difference between the two. This enables a quantitative comparison of tumor images and ABM simulations, where the distance between simulated and experimental images can be minimized using standard parameter-fitting algorithms. Here, we describe this method and present two examples to demonstrate the application of the approach to estimate parameters for two distinct ABMs. Overall, we provide a novel method to robustly estimate ABM parameters.Author summary: Parameter estimation is a key step in computational model development, and accurate parameters are required to produce robust model predictions. Agent-based models (ABMs) are commonly used to simulate tumor growth; however, these models are exceedingly difficult to fit to experimental or clinical imaging data due to the complex spatial relationships of various cell types. Currently, simple comparison metrics extracted from tumor images and ABM simulations are used to qualitatively assess the model fit. In this work, we present a novel method for comparing spatial ABM simulations to tumor images as a single quantitative value that measures how different the two are and can then be used as the objective function for a parameter estimation algorithm. Our approach uses representation learning, where a neural network is used to project an input into low-dimensional space. This method can be used to aid researchers in developing and fitting tumor ABMs based on actual patient data.