SNP Feature Selection Using LD Filtering and Heritability Screening

The patent extracts and removes redundant SNP data through multiple filtering stages. The LD-based filtering extracts and removes SNPs in linkage disequilibrium with already-selected SNPs, while the variance-based filtering extracts and removes SNPs with insufficient variability. This extraction of redundant information reduces data complexity while preserving essential genetic signals.

The patent segments the SNP selection process into distinct stages: initial LD-based filtering, variance-based filtering, and heritability-based selection. Each segment handles a specific aspect of data reduction, breaking down the complex task of selecting core SNPs from tens of thousands of candidates into manageable sequential steps that collectively achieve both completeness and efficiency.

The patent performs preliminary actions by conducting LD-based filtering and variance-based filtering before the final heritability-based SNP selection. These preliminary filters remove obviously redundant SNPs (those in LD with others or with low variance) while preserving SNPs that have potential heritability value, ensuring that important genetic information is retained for the subsequent heritability analysis stage.

The patent incorporates feedback mechanisms where each filtering stage uses information from previous stages to guide subsequent selection. The LD-based filtering provides feedback on which SNPs are redundant, the variance filtering provides feedback on which SNPs lack variability, and both feed into the final heritability-based selection that uses this accumulated information to make informed decisions about core SNP selection.

The patent segments the feature extraction process into three distinct methods applied in sequence: LD-based filtering, variance-based filtering, and heritability-based selection. Each method focuses on a specific aspect of SNP characterization, allowing the system to achieve high selection accuracy through cumulative refinement rather than requiring one computationally intensive method to handle all aspects simultaneously.

The patent applies partial filtering actions at each stage, removing only the most obviously redundant SNPs in early stages (LD and variance filtering) before applying the more computationally intensive heritability-based selection to a reduced dataset. This partial action approach achieves sufficient accuracy for preliminary filtering without the full computational cost of applying heritability analysis to all SNPs.