A benchmark study of scoring methods for non-coding mutations.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Pathogenic variant prioritization is crucial for genetic disease diagnosis.
• While coding regions have established prioritization models, non-coding regions lack comprehensive performance assessments for deleteriousness scores.

Purpose of the Study:

• To conduct a large-scale comparison of leading and recent deleteriousness scoring tools for non-coding variants.
• To evaluate the performance of these tools in discriminating pathogenic from benign variants using established databases.

Main Methods:

• Compared CADD, FATHMM-MKL, Funseq2, GWAVA, DANN, SNP, and SOM scores.
• Utilized ClinVar, COSMIC, and 1000 Genomes Project databases for benchmark comparisons.
• Assessed the ability of tools to discriminate pathogenic from benign variants.

Main Results:

• CADD excelled in identifying pathogenic variants in coding gene regions using the ClinVar benchmark.
• FATHMM-MKL, GWAVA, and SOMliver showed superior performance for non-coding variants (lincRNAs, pseudogenes) using the COSMIC benchmark.
• All tools exhibited low precision, highlighting the need for novel non-coding genome feature discoveries.

Conclusions:

• The performance of non-coding variant prioritization tools varies depending on genomic location (coding vs. non-coding).
• Current tools have limitations in precision, necessitating further research into non-coding genome features.
• Development of a robust gold standard dataset for non-coding regions is essential for accurate tool evaluation.