NCBoost classifies pathogenic non-coding variants in Mendelian diseases through supervised learning on purifying...

Area of Science:

• Genetics
• Bioinformatics
• Genomic Medicine

Background:

• Current methods for assessing pathogenic non-coding variants have limitations in accuracy and positional bias.
• These methods are often evaluated on common disease polymorphisms, not specifically monogenic diseases.

Purpose of the Study:

• To develop a more accurate and less biased method for identifying pathogenic non-coding variants.
• To curate a high-confidence set of pathogenic non-coding variants for training and testing.

Main Methods:

• Curated 737 high-confidence pathogenic non-coding variants for Mendelian diseases.
• Incorporated interspecies conservation and human purifying selection signals.
• Utilized gradient tree boosting (NCBoost) for supervised learning.

Main Results:

• Achieved high predictive performance in identifying pathogenic non-coding variants.
• Overcame the positional bias inherent in previous methods.
• NCBoost demonstrated consistent performance across different datasets.

Conclusions:

• NCBoost offers a significant advancement in the accurate identification of pathogenic non-coding variants.
• The method outperforms existing reference approaches.
• This tool has implications for diagnosing Mendelian diseases.