NPDR

NPDR detects important predictors and interaction effects in high-dimensional biological datasets by projecting nearest-neighbor distances onto predictor dimensions and assessing significance with generalized linear model regression.

Key Features:

• Nearest-Neighbor Approach: NPDR leverages a nearest-neighbor framework to capture the interaction structure within high-dimensional datasets by calculating distances between nearest-neighbor pairs projected onto predictor dimensions.
• Generalized Linear Model Regression: NPDR employs generalized linear model regression to evaluate predictor significance for dichotomous and continuous outcomes across varied data types.
• Covariate Adjustment and Multiple Testing Correction: NPDR adjusts for covariates and applies multiple testing correction to reduce false positives.
• Statistical Inference and Penalized Regression: NPDR supports statistical inference and includes penalized regression capabilities for high-dimensional feature selection.

Scientific Applications:

• GWAS and eQTL Studies: NPDR can identify interacting genetic variants that regulate transcripts in GWAS and eQTL studies, exemplified by analyses related to major depressive disorder (MDD).
• Gene Expression Analysis: Applied to RNA-Seq data, NPDR can discern gene expression patterns and interactions relevant to disease mechanisms.
• Neuroimaging Studies: In structural and functional neuroimaging, NPDR can uncover network interactions underlying cognitive or behavioral phenotypes.

Methodology:

NPDR projects distances between nearest-neighbor pairs onto predictor dimensions and fits generalized linear model regression to assess predictor importance; simulations reported improved precision-recall performance relative to standard Relief-based methods and random forest importance metrics and demonstrate enhanced ability to detect interactions and other effects.