Unveiling the Hidden Rules: Enhancing NMD Prediction for Protein-Truncating Variants

Area of Science:

• Genetics
• Molecular Biology
• Bioinformatics

Background:

• Nonsense-mediated decay (NMD) is a crucial RNA quality control pathway.
• Premature termination codons (PTCs) trigger NMD, degrading aberrant transcripts.
• Accurate prediction of NMD is vital for interpreting genetic variants, especially those causing truncated proteins.

Purpose of the Study:

• To develop a predictive model for NMD sensitivity of transcripts with PTCs.
• To improve upon the limitations of the canonical 50-55 nucleotide rule in predicting NMD outcomes.
• To create a scalable resource for interpreting the functional impact of protein-truncating variants.

Main Methods:

• Utilized paired whole-genome and RNA-sequencing data from 10,306 individuals in the Trans-Omics for Precision Medicine (TOPMed) program.
• Quantified NMD efficiency for 5,749 germline truncating variants using allele-specific expression analysis.
• Trained a gradient-boosting classifier (TrunCat) incorporating features like SHAP values to predict NMD sensitivity.

Main Results:

• The TrunCat classifier achieved approximately 78% ROC-AUC in distinguishing NMD-sensitive from NMD-escape transcripts.
• A simplified model using top SHAP features demonstrated comparable predictive performance.
• Application to variant databases and a rare-disease cohort showed differential NMD predictions for pathogenic versus uncertain significance variants.

Conclusions:

• The developed TrunCat model significantly enhances the prediction of NMD outcomes for PTC-containing transcripts.
• This framework validates the canonical NMD rule while uncovering novel determinants of NMD sensitivity.
• Provides a valuable, scalable tool for genetic variant interpretation and understanding disease associations.