BANNs

BANNs model genotype–phenotype associations in genome-wide association (GWA) studies using partially connected feedforward neural networks that integrate biological annotations to map SNP-level effects to SNP-sets and perform enrichment analysis.

Key Features:

• Nonlinear Probabilistic Framework: BANNs employ nonlinear feedforward neural network architectures to capture non-additive genetic variation beyond traditional linear models.
• Interpretable Architecture: The input layer encodes single nucleotide polymorphism (SNP)-level effects while hidden units aggregate these effects within annotated SNP-sets (e.g., genes, signaling pathways) via partially connected layers.
• Probabilistic Inference: Network weights and connections are treated as random variables with prior distributions, and variational inference is used to produce posterior summaries for SNP-level associations and SNP-set enrichment.
• Simultaneous Mapping and Enrichment: The framework jointly performs SNP-level association mapping and enrichment analysis of SNP-sets such as genes or signaling pathways.

Scientific Applications:

• Association Mapping in GWA Studies: BANNs are applied for association mapping in GWA studies and have demonstrated superior performance compared to state-of-the-art methods across diverse genetic architectures.
• Lipid Trait Association: BANNs have been applied to high- and low-density lipoprotein cholesterol (HDL and LDL) association analyses.
• Large Cohort and Biobank Analyses: BANNs have been used on GWA datasets from the Wellcome Trust Centre for Human Genetics, the Framingham Heart Study, and the UK Biobank to replicate known associations and identify novel genomic links.

Methodology:

Partially connected feedforward neural networks integrate biological annotations with the input layer encoding SNP-level effects and the hidden layer aggregating SNP-sets; network weights and connections are assigned prior distributions and approximate Bayesian inference via variational inference is used to obtain posterior summaries for association mapping and enrichment analysis.