Improving variant prioritization in exome analysis by entropy-weighted ensemble of multiple tools

Area of Science:

• Genomics
• Bioinformatics
• Medical Genetics

Background:

• Variant prioritization is essential for analyzing exome and genome sequencing data in clinical settings.
• Existing phenotype-driven tools face challenges with imprecise phenotypic information and their comparative efficacy is not well-established.

Purpose of the Study:

• To assess the efficacy of individual variant prioritization tools (Exomiser, Xrare, DeepPVP) and their ensemble in a large rare disease cohort.
• To determine if an ensemble approach can outperform single algorithms, especially with heterogeneous phenotypic data.

Main Methods:

• Evaluation of three freely available variant prioritization tools: Exomiser, Xrare, and DeepPVP.
• Development and assessment of an ensemble method using a weighted-sum entropy approach (EWE3).
• Validation of findings in a primary cohort (1614 patients) and a replication cohort (1904 patients).

Main Results:

• The performance of individual tools was sensitive to the quality of phenotypic input.
• The entropy-weighted ensemble (EWE3) achieved 78% diagnostic variants in the top 3, a 13% improvement over the best single tool.
• The ensemble method reached 88% in top 10 and 96% in top 30, with results replicated in an independent cohort.

Conclusions:

• An entropy-weighted ensemble of multiple variant prioritization tools significantly enhances diagnostic accuracy in exome and genome sequencing.
• This ensemble approach is effective in clinical settings, even with complex and imprecise phenotypic information.
• Combining tools like Exomiser, Xrare, and DeepPVP via ensemble methods accelerates molecular diagnosis for rare diseases.