Predicting Translation Readthrough from Sequence Features in Saccharomyces cerevisiae

Understanding translation readthrough in eukaryotes is fundamental for elucidating transcript evolution, deciphering gene expression codes, and advancing biotechnological applications.In this study, we present the first comprehensive analysis of the transcript features that determine readthrough efficiency in Saccharomyces cerevisiae.To predict readthrough efficiency, we engineered a set of novel sequence-derived features and trained multiple machine-learning models, achieving a mean Spearman correlation of 0.535 between predicted and experimentally measured values, substantially improving predictive performance compared with models based solely on previously established features.Feature importance analysis identified codon usage-related features, particularly the tRNA Adaptation Index (tAI), Codon Adaptation Index (CAI), and similar-to-stop codon counts, as the strongest contributors to model performance.Lower tAI values near the stop codon were strongly associated with increased readthrough, indicating that reduced translational efficiency promotes termination errors.In contrast, higher similar-to-stop codon counts were positively associated with readthrough, supporting the hypothesis that local tRNA competition increases the likelihood of stop codon misreading.Together, these findings highlight the central role of codon usage and local sequence context in modulating translation termination fidelity.This work provides a predictive framework for readthrough that can facilitate future studies aimed at understanding and modulating translation termination.