TREAT

TREAT implements tree-structured regression via recursive partitioning in R to detect complex joint effects among multiple predictors and disease outcomes in case-control studies.

Key Features:

• Tree-Structure Model: Employs a tree-structure model to capture non-additive and interaction effects among predictors in genetic analyses.
• Recursive Partitioning: Constructs tree-structured models through recursive partitioning of the data.
• Ultra-fast Algorithm: Uses an ultra-fast algorithm to build and evaluate tree-structured models efficiently, reducing computational time for hypothesis testing.
• Adaptive Model Selection: Implements an adaptive model selection procedure to identify the optimal tree model representing joint effects.
• Statistical Performance: Demonstrates superior power relative to other commonly used tests in simulation studies.
• Alternative to Logistic Regression-based Multivariate Tests: Provides a tree-based alternative to traditional multivariate tests derived from logistic regression models.
• Computational Implementation: Implemented in C++ and R.

Scientific Applications:

• Multilocus Association Testing: Applied as a multilocus association test on over 20,000 genes/regions to assess joint effects of multiple loci.
• Genetic Epidemiology: Used in genomics and epidemiology to detect complex gene-disease interactions in case-control studies.
• Discovery of Disease Associations: Enabled identification of a highly significant association between CDKN2B and esophageal squamous cell carcinoma (ESCC) in applied studies.

Methodology:

Constructs tree-structured models via recursive partitioning, applies an adaptive model selection procedure, and evaluates significance using an ultra-fast algorithm to mitigate computational challenges of tree-based hypothesis testing.