Deep learning model for predicting mRNA half-life based on 3'UTR sequences.

Area of Science:

• Computational Biology
• Molecular Biology
• Genetics

Background:

• Messenger RNA (mRNA) stability, determined by its half-life, is crucial for gene expression regulation.
• Predicting mRNA half-life solely from 3' untranslated region (3'UTR) sequences is complex, demanding both accuracy and biological insight.

Purpose of the Study:

• To develop a sequence-driven computational framework for predicting mRNA half-lives in Saccharomyces cerevisiae.
• To integrate a pre-trained RNA language model (RNA-FM) with a Transformer architecture for enhanced prediction accuracy and biological interpretability.

Main Methods:

• Utilized a pre-trained RNA language model (RNA-FM) combined with a Transformer backbone.
• Applied the framework to predict mRNA half-lives from 3'UTR sequences of Saccharomyces cerevisiae transcripts.
• Employed within-gene evaluation focusing on novel isoforms and in-silico mutagenesis for motif identification.

Main Results:

• Achieved an RMSE of 8.76 min and MAE of 5.95 min on an independent test set, with R² of 0.545.
• Identified specific GU-enriched motifs (e.g., UUGUAU, AUGCA) as destabilizing elements and others (e.g., UUUAUG, GAGGU) as stabilizing.
• Demonstrated robust positional dependencies for these motifs and minimal systematic bias in predictions.

Conclusions:

• The developed framework accurately predicts mRNA half-lives and provides mechanistic insights into post-transcriptional regulation.
• Identified key sequence motifs in 3'UTRs that influence mRNA stability, offering a practical tool for studying mRNA homeostasis.
• The approach facilitates a deeper understanding of gene expression regulation and can guide future mRNA design strategies.