Predicting Microplastics Exposure via Neural Network Gene Expression Integration

Overview

This study investigated the predictability of microplastic exposure in Daphnia magna through a machine learning-based approach based on gene expression levels. The research's methodology involved training the model for 50 epochs on a dataset containing gene expression levels for different Daphnia magna clones exposed to microplastics. Another model essentially simulated a dataset of Daphnia magna not exposed to microplastics, predicting the likelihood of the control group's exposure. The independent variable was gene expression levels, while the dependent variable was the model's accuracy in predicting microplastic exposure, evaluated through metrics like accuracy or loss. The latter has shown discernible patterns while predicting microplastic exposure per every target gene and model chosen. Specifically, BL2.2, Max4, and K34J models were accurate with some target genes showing perfect accuracy. The ROC curves demonstrated excellent performance of those models during the binary classification test. These results supported the hypothesis and emphasized the model's predictive ability on microplastic exposure in Daphnia magna, displaying its effectiveness.