fastJT

fastJT implements the Jonckheere-Terpstra statistic as an R package for robust, rank-based feature selection in high-dimensional genomic studies (including GWAS) and machine learning, accommodating scenarios where genotypes have assumed additive effects on phenotypes.

Key Features:

• Robustness to Outliers: A rank-based, distribution-free Jonckheere-Terpstra implementation reduces sensitivity to outliers and influential observations.
• Computational Efficiency: Replaces pairwise comparison and counting with a data sorting and searching procedure, reducing computational complexity from O(n^2) to O(n log n).
• Parallel Processing: Shared-memory parallel programming enables leveraging multi-core processors to accelerate computations.
• Application to Constrained Hypotheses: Supports analysis of ordered predictors or genetic variants where natural ordering or assumed additive genotype effects on phenotypes are expected.

Scientific Applications:

• Genome-Wide Association Studies (GWAS): Facilitates identification of genetic variants associated with phenotypic traits in large-scale GWAS datasets using the Jonckheere-Terpstra statistic.
• Machine Learning Feature Selection: Provides a statistical criterion to evaluate and select features that follow monotonic or ordering constraints for machine learning models.

Methodology:

Computes the Jonckheere-Terpstra statistic using an algorithm that leverages internal sample information to avoid unnecessary comparisons, replaces pairwise comparisons and counting with sorting-and-searching to achieve O(n log n) complexity, and employs shared-memory parallelization.