Machine learning evaluating evolutionary fitness in complex biological systems

• Main Authors: Kuzenkov, Oleg, Morozov, Andrew, Kuzenkova, Galina
• Format: Conference Proceeding
• Language: English
• Subjects: Analytical models; Biological system modeling; diel vertical migration; evolutionarily stable strategy; fitness; machine-learned ranking; neural network; Neural networks; recognition; Sociology; Software; Time series analysis; validation; zooplankton
• ISSN: 2161-4407
• DOI: 10.1109/IJCNN48605.2020.9206653

Here we suggest a novel computational approach based on artificial neural network technologies to be able to evaluate evolutionary fitness in both theoretical models of population dynamics and empirical biological systems from data. Our approach uses long-time population time series (obtained either from a model or from data) and establishes the ranking order of inherited strategies reflecting their selective advantages. We approximate the fitness surface in the space of a few key parameters based on Taylor expansion. To do this, we create learning and testing samples and then apply artificial neural networks to build a fitness surface separating the domains of interior and superior ranking in the space of parameters. Using the obtained approximation of the fitness function we can find the evolutionarily stable (optimal) strategy by maximising evolutionary fitness. We demonstrate the efficiency of our approach by applying it to some classical population models where the exact fitness function can be derived analytically as well as to empirical systems. In the considered study cases, both the fitness function and the optimal strategy obtained via our computational method are close to the ones provided by analytical solutions or observed in natural systems. We apply our method to predict the evolutionary stable diel vertical migrations (DVM) of zooplankton in the ocean and lakes, the phenomenon, which is considered as the most significant synchronous biomass movement on Earth.