Precise uncertain significance prediction using latent space matrix factorization models: genomics variant and...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Existing variant interpretation methods struggle with the large number of genetic variants and variants of uncertain significance (VUS).
• Accurate prediction of disease type and disease-gene associations is crucial for clinical genetics.
• Current approaches often focus on individual variants, leading to challenges in classifying overall gene pathogenicity.

Purpose of the Study:

• To develop algorithms for assigning a degree of pathogenicity to genes rather than individual variants.
• To address the challenge of classifying variants of uncertain significance (VUS) in genomic data.
• To improve the accuracy of disease-gene association predictions by focusing on gene-level significance.

Main Methods:

• Proposed algorithms to assign gene significance scores based on pathogenicity.
• Designed two matrix factorization-based models: a common latent space model (using genomic and clinical data) and a single-matrix factorization model (for scenarios with limited clinical data).
• Evaluated model performance by training five multi-label classifiers, including a feedforward neural network, on novel input features.

Main Results:

• The proposed models successfully predict uncertain significance scores with high accuracy and low error.
• The novel input features significantly contribute to the high accuracy achieved by the multi-label classifiers.
• The common latent space model effectively integrates genomics variant data and heterogeneous clinical data for improved predictions.

Conclusions:

• The developed gene-centric approach effectively addresses limitations in variant interpretation and VUS classification.
• Matrix factorization models provide accurate predictions for gene pathogenicity, enhancing disease-gene association studies.
• The approach offers a robust framework for understanding gene significance in the context of human diseases.