VEPAD

VEPAD predicts the impact of single nucleotide variants (SNVs) associated with Alzheimer's disease, classifying them as deleterious or neutral to support variant interpretation in AD research.

Key Features:

• Data Integration: Training data comprise 20,401 deleterious variants and 37,452 control variants sourced from Genome-Wide Association Study (GWAS) and Genotype-Tissue Expression (GTEx) databases.
• Input Features: Models incorporate seven histone marks from various brain tissues, two conservation scores, GC content, and CpG site characteristics around mutation sites.
• Feature Selection: Recursive Feature Elimination with Cross-Validation (RFECV) followed by forward feature selection is used to identify critical genomic and epigenomic markers.
• Machine Learning Model: A Random Forest classifier is employed, achieving 81.21% accuracy on 10-fold cross-validation and 70.63% accuracy on an independent test set of 5,785 variants.
• Comparative Performance: Compared to existing tools such as CADD and FATHMM, which achieve reported accuracies of 54%–62%, VEPAD attains higher accuracy on the evaluated datasets.

Scientific Applications:

• Early Diagnosis: Identification of deleterious AD-associated variants that can inform early genetic risk assessment.
• Biomarker Discovery: Prioritization of variants and associated epigenomic markers for development of targeted biomarkers and therapeutic strategies.
• Genetic Research: Systematic evaluation of variant effects to investigate genetic contributions to AD pathogenesis and progression.

Methodology:

VEPAD integrates genomic and epigenomic features (seven brain histone marks, two conservation scores, GC content, CpG characteristics) from GWAS and GTEx, applies RFECV and forward feature selection, trains a Random Forest classifier, and evaluates performance by 10-fold cross-validation and on an independent test set of 5,785 variants.