Field-Testing a Proteomics-Derived Machine-Learning Model for Predicting Coral Bleaching Susceptibility

Abstract

Given the widespread decline of coral reefs, temperature-focused models have been generated to predict when and where bleaching events may occur (e.g., Coral Reef Watch). Although such algorithms are adept at forecasting the onset of bleaching in many areas, they suffer from poor predictive capacity in regions featuring corals that have adapted or acclimatized to life in marginal environments, such as reefs of the Florida Keys (USA). In these locales, it may instead be preferred to use physiological data from the corals themselves to make predictions about stress tolerance. Herein proteomic data from both laboratory and field samples were used to train neural networks and other machine-learning models to predict coral bleaching susceptibility in situ, and the models’ accuracies were field-tested with massive corals (Orbicella faveolata) sampled across a 2019 bleaching event. The resulting artificial intelligence was capable of accurately predicting whether or not a coral would bleach in response to high temperatures based on its protein signatures alone, meaning that this approach could consequently be of potential use in delineating O. faveolata climate resilience.