Impact of mRNA modifications induced by glycemic stress on the efficiency of protein translation

Stochastic Translation Efficiency (STE) is a key indicator for studying cellular adaptation mechanisms, but measuring it experimentally remains complex.The objective of this work is to construct a predictive model of STE using machine learning methods that integrate multi-omic data in yeast Saccharomyces cerevisiae. The data analysed covers approximately 5,000 genes and combines several types of explanatory variables: structural characteristics, messenger RNA modifications and degradation measurements. The observations are available under two experimental conditions: a control condition and a condition subjected to 10 minutes of glycaemic stress.The model is trained using data from the control condition. It is based on a three-step architecture: (i) dimension reduction to address the high correlation and redundancy of explanatory variables, (ii) a clustering method to identify gene groups, and (iii) a regression model to predict STE.This integrated approach addresses the complexity and high dimensionality of multi-omic data, while improving the stability and robustness of translation efficiency predictions.