Agent-Based Modeling and Machine Learning to Explore Multi-cell Spheroid Patterning

Dynamically activated differential adhesion between cell subpopulations drives multicellular tissue patterning in development and disease. Previous studies have explored this process in heterogeneous spheroids by using synthetically engineered systems, in which initially non-adherent cells engage in bi-directional signaling to activate differential cadherin expression. While synthetic systems provide an excellent in vitro model to observe pattern formation, computational techniques can be leveraged to systematically explore the key parameters that drive the emergence of different patterns. We developed and validated two- and three-dimensional computational agent-based models (ABMs) of cell patterning in spheroids and demonstrated how varying the initial cell seeding ratio, signaling sensitivity, and homotypic adhesion strengths between cells lead to unique spheroid patterns. We developed novel model exploration techniques that use machine learning to identify how combinations of cell-to-cell signaling parameters in this system drive the formation of specific multicell patterns. We also deployed the model in reverse as a tool to design new synthetic cell signaling circuits based on a desired final multicell pattern.