Agent-based modelling and time series inference of filamentous yeast colonies

Abstract

The bakers yeast Saccharomyces cerevisiae can form invasive filamentous colonies with non-uniform spatiotemporal patterns. We aimed to better understand how individual cellular actions give rise to colony-scale patterns. We used an off-lattice agent-based model to simulate colony growth, and used a neural likelihood estimator (NLE) to infer parameters for experimental photographs. Li et al. [1] used approximate Bayesian computation (ABC) to infer parameters using coarse summary statistics obtained from a single time point averaged across experimental replicates. The NLE overcomes the computational expense of ABC, allowing us to infer the parameters of individual colonies from a full time series of experimental photographs. To demonstrate the capabilities of our model and inference technique, we tested extensively on synthetic data and then predicted yeast growth under three different experimental conditions. As before, the proportion of total colony growth above which pseudohyphal growth is permitted was a key parameter that contributed to colony morphology. Since our NLE-based approach incorporates time series data, it yielded better parameter estimates and more accurate predictions compared to our ABC-based method. This updated approach improved understanding of how the probability that a stated cell produces a pseudohyphal cell influences colony morphology. In this way, the model also has the potential to generate hypotheses, which can be tested through biological experiments to increase the understanding of the basis for different growth patterns in yeast.