From mutation to degradation: predicting nonsense-mediated mRNA decay with NMDap

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Nonsense-mediated mRNA decay (NMD) is a vital surveillance pathway that degrades aberrant transcripts.
• Accurate prediction of NMD activity is challenging due to limitations in current models.
• NMD influences transcriptome integrity and disease phenotypes.

Purpose of the Study:

• To develop an accurate and generalizable predictor for NMD activity.
• To identify key determinants of NMD using explainable AI.
• To enable large-scale assessment of mRNA degradation for variant interpretation.

Main Methods:

• Benchmarking of embedding-only models against rule-based approaches using TCGA and GTEx data.
• Development of NMDap, an integrative framework combining rule-based methods, sequence embeddings, and biological features.
• Application of explainable AI to identify NMD determinants.

Main Results:

• Embedding-only models showed underperformance compared to simple rule-based methods.
• NMDap achieved improved predictive performance over existing models.
• Key NMD determinants were identified, including mean ribosome loading.

Conclusions:

• NMDap offers a robust framework for predicting NMD activity and understanding its regulation.
• The tool facilitates large-scale mRNA degradation assessments, aiding variant interpretation and disease research.
• Explainable AI enhances the biological insights derived from NMD prediction models.