Agent-Based Simulations of Lung Tumor Evolution Suggest That Ongoing Cell Competition Drives Realistic Clonal Expansions.

Computational simulations of tumor evolution are increasingly used to infer the rules underlying cancer growth. To make reliable inferences, such models must be able to reflect the properties of real tumors. Recent work has shown that lung tumors undergo frequent and late subclonal expansions, which are associated with poor prognosis. This paper tests three candidate simulations of three-dimensional tumor growth, which make different assumptions about the nature of competition between cells, for their ability to replicate these late expansions. The study identifies a computationally-efficient model which can produce multi-region sequencing data realistic to lung tumors. This model assumes two distinct stages of growth, with the second stage involving local competition for space and resources within and between small tissue areas in a fixed-size tumor. When inferring the model-specific fitness effect of driver mutations in a large cohort of lung cancers, the study finds that inference pipelines based on a two-stage model imply much larger selection effects than those based on single-stage models, driven by model-specific assumptions about the practical consequences of selection strength. This work emphasizes the importance of model assumptions to the results of tumor-specific, simulation-based inferences.