Deep learning approach to the estimation of the ratio of reproductive modes in a partially clonal population

TL;DR

This paper introduces a deep learning model to estimate the ratio of sexual to asexual reproduction in populations with mixed reproductive strategies.

Contribution

The novel contribution is a convolutional neural network model that accurately estimates reproductive mode ratios in partially clonal populations.

Findings

The model achieves an accuracy of up to 0.99 in estimating reproductive mode ratios.

The model performs better when training data dimensions match the actual data.

The approach is suitable for neutral multiallelic marker traits like microsatellite repeats.

Abstract

Genetic diversity among biological entities, including populations, species, and communities, serves as a fundamental source of information for understanding their structure and functioning. However, many ecological and evolutionary problems arise from limited and complex datasets, complicating traditional analytical approaches. In this context, our study applies a deep learning-based approach to address a crucial question in evolutionary biology: the balance between sexual and asexual reproduction. Sexual reproduction often disrupts advantageous gene combinations favored by selection, whereas asexual reproduction allows faster proliferation without the need for males, effectively maintaining beneficial genotypes. This research focuses on exploring the coexistence patterns of sexual and asexual reproduction within a single species. We developed a convolutional neural network model specifically designed to analyze the dynamics of populations exhibiting mixed reproductive strategies within changing environments. The model developed here allows one to estimate the ratio of population members who originate from sexual reproduction to the clonal organisms produced by parthenogenetic females. This model assumes the reproductive ratio remains constant over time in populations with dual reproductive strategies and stable population sizes. The approach proposed is suitable for neutral multiallelic marker traits such as microsatellite repeats. Our results demonstrate that the model estimates the ratio of reproductive modes with an accuracy as high as 0.99, effectively handling the complexities posed by small sample sizes. When the training dataset’s dimensionality aligns with the actual data, the model converges to the minimum error much faster, highlighting the significance of dataset design in predictive performance. This work contributes to the understanding of reproductive strategy dynamics in evolutionary biology, showcasing the potential of deep learning to enhance genetic data analysis. Our findings pave the way for future research examining the nuances of genetic diversity and reproductive modes in fluctuating ecological contexts, emphasizing the importance of advanced computational methods in evolutionary studies.

Figures