A General Framework for Designing Evolutionary Experiments to Select Specific Phage Phenotypes Using Neural Networks, Statistical Simulations, and Symbolic Regression.

Understanding how environmental conditions shape the evolution of bacteriophages (phages) is critical for designing correct evolutionary experiments that select specific phage traits. This study provides a general mathematical framework that integrates physics-informed neural networks, agent-based statistical simulations, and symbolic regression machine learning techniques to design evolutionary experiments targeting specific phage traits such as high variability in phage phenotypes.

In the study, we used agent-based statistical simulations to generate synthetic time series data for evolutionary scenarios with diverse phenotypic outcomes. Subsequently, we trained Physics-informed neural networks (PINNs) embedded in differential equations on the synthetic time series to reveal possible environments that select given phage traits and uncovered hidden interactions in the system [1].

Lastly, by using symbolic regression techniques, including Sparse Identification of Nonlinear Dynamical Systems (SINDy) and genetic algorithm-based methods, on the synthetic time series data, we derived governing differential equations that identified key environmental conditions and interaction pathways driving the evolution of specific phage traits in well-mixed and spatiotemporal microbial systems. These equations serve as a foundation for predicting the correct evolutionary experiment to select for the given phage traits.

This approach provides a scalable and adaptable framework for designing evolutionary experiments to select specific phage phenotypes. It highlights the power of integrating physics-informed neural networks, agent-based simulations, and symbolic regression techniques to offer new insights into the mechanisms underlying complex microbial ecosystems.

Reference:

[1] Grigorian, G., George, S.V. and Arridge, S., 2024. Learning Governing Equations of Unobserved States in Dynamical Systems. arXiv preprint arXiv:2404.18572.