Predicting functional UTR variants by integrating region-specific features.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Untranslated regions (UTRs) of mRNA regulate gene expression, and variants within them are linked to human diseases.
• Computational prediction of UTR variant effects is challenging, with current methods often neglecting UTR-specific features.

Purpose of the Study:

• To develop accurate computational models for predicting the functional impact of UTR variants.
• To identify key sequence determinants driving functional UTR variants.

Main Methods:

• Systematic analysis of over 50 region-specific UTR features using consolidated variant datasets.
• Development of machine learning classification models utilizing identified UTR features.

Main Results:

• Identified sequence composition features (e.g., C/G in 5'UTR, A/T in 3'UTR) that differentiate functional from non-functional variants.
• Achieved high predictive performance with AUC values of 0.94 for 5'UTR and 0.85 for 3'UTR.
• Developed models that outperform existing methods for UTR variant prediction.

Conclusions:

• The developed machine learning models significantly enhance the prediction of functional UTR variants.
• These models offer valuable tools for clinical interpretation of genetic variants and disease risk assessment.