Tracking the frequency of phytoplankton clonal lineages using multispectral image flow cytometry and neural networks

Abstract

Abstract Testing fundamental theories of diversity and evolutionary change often requires tracking the frequencies of clonal lineages within a population over time. Current tracking methods in plankton and microbial systems are often labour‐intensive, time‐consuming, expensive and/or unavailable, especially when high temporal resolution of frequencies is required. The combination of multispectral imaging flow cytometry and neural networks (NN) could provide an efficient approach to classify clonal lineages by their heritable phenotypes and thus track their frequency changes. Here, we present a novel method that combines NN and feature values extracted from images to classify six clonal lineages of the green alga Chlamydomonas reinhardtii based on heritable morphological differences when paired in‐silico and in‐vitro. We compared different NN trained on all six clonal lineages or on pairs of two and compared the accuracy of the models. All NN were able to classify clonal lineages with very high accuracy and the NN trained on pairs of clonal lineages achieved the highest accuracy of 97%. Using in‐vitro samples we observed a drop in accuracy to an average of 85%, which was due to the variation occurring between image acquisitions and culturing conditions causing differences in the extracted feature values. This method can be applied to reduce the workload and running costs of long‐term lineage tracking, and it can be applied to a wide range of plankton and microbial organisms and research questions.

Keywords

• image flow cytometry
• machine learning
• multispectral imaging
• neural networks
• phytoplankton