Transition-transversion encoding and genetic relationship metric in ReliefF feature selection improves pathway...

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Statistical Genomics

Background:

• ReliefF is a nearest-neighbor algorithm for feature selection, identifying important genetic variants through statistical interactions (epistasis).
• Traditional ReliefF uses a simple mismatch difference for categorical predictors like genotypes.
• Existing methods do not fully capture complex genetic relationships or allele frequency variations.

Purpose of the Study:

• To develop and evaluate novel metrics for the ReliefF algorithm.
• To incorporate allele sharing, genetic relationship matrix (GRM) adjustment, and transition/transversion encoding into ReliefF.
• To enhance the detection of biologically relevant genetic variants in genome-wide association studies (GWAS).

Main Methods:

• Introduced a novel two-dimensional transition/transversion genotype encoding for ReliefF.
• Implemented and compared three attribute metrics: genotype mismatch (GM), allele mismatch (AM), and transition/transversion.
• Integrated GRM as a nearest-neighbor metric to account for allele frequency heterogeneity.

Main Results:

• Applied ReliefF with new metrics to a GWAS for major depressive disorder.
• Demonstrated improved detection of genes within depression-implicated pathways (e.g., Axon Guidance, G Protein-Coupled Receptor Signaling).
• Compared performance against Random Forest, Lasso, and random selection to control for pathway size bias.

Conclusions:

• Genetically informed encodings (transition/transversion) and allele frequency adjustment (GRM) significantly improve ReliefF attribute scores.
• These enhanced metrics lead to superior pathway enrichment in genetic association studies.
• The refined ReliefF approach offers greater power for discovering complex genetic underpinnings of diseases.