Resumen en inglés

Coral reefs are key contributors to marine biodiversity, and their survival depends on the symbiotic relationship between corals and Symbiodiniaceae algae. One of the main natural dispersers of these microalgae is the parrotfish, which releases them through its feces. Therefore, identifying which cells in fecal samples correspond to Symbiodiniaceae, and more importantly, determining their viability, is crucial for understanding coral reef ecosystems. Flow cytometry is a powerful tool for analyzing cellular characteristics at the single-cell level, but manual classification of these cells is time-consuming and subject to human bias. This study explores the use of machine learning (ML) to automate the classification of algae and sediment particles in parrotfish fecal samples. Initial clustering was carried out using unsupervised algorithms (K-Means and Gaussian Mixture Models), with K-Means producing clearer separation in PCA-reduced feature space. These clusters were used to train supervised models, including Support Vector Machines (SVM), Random Forest and HistGradientBoosting. While models performed similarly, HistGradientBoosting was selected due to its faster training time and comparable accuracy. The models successfully captured biologically meaningful patterns, including the gradual transition from viable to non-viable cells and the accurate identification of sediment particles. This study highlights the value of combining flow cytometry with machine learning to support ecological research and conservation of coral reef environments.