CombInference Package

Overview

CombInference is an R package designed to implement a novel framework for combinatorial inference in graphical models, with False Discovery Rate (FDR) control across multiple scales. The package applies methods introduced in the paper:

Lu Zhang. Junwei Lu. "StarTrek: Combinatorial variable selection with false discovery rate control." Ann. Statist. 52 (1) 78 - 102, February 2024. https://doi.org/10.1214/23-AOS2296.

The core functionality of this package includes:

1. Graphical feature selection.
2. Persistent homology analysis with uniform FDR (uFDR) control.
3. Integration of the KHAN algorithm for homological feature selection.

Key Features

• Graphical Feature Selection: Implementing selection of graph features like hubs, cycles, and edges, based on single-edge p-values with FDR control.
• Persistent Homology: Efficient analysis of persistent homological features in graphs, across continuous filtration levels.
• KHAN Algorithm: Adaptive selection of homological features using a discrete Gram-Schmidt procedure to ensure statistical and computational efficiency.

Installation

To install the CombInference package, run:

install.packages("CombInference")

Usage

# Example call to the moonshoot function:
result <- moonshoot(
  X = your_data_matrix,
  SigmaHat = your_covariance_matrix,
  ThetaHat = your_precision_matrix,
  q = 0.05,
  numB = 1000,
  V0 = your_vertex_matrix,
  possibleSet = your_possible_set_list,
  prescreen = TRUE
)

---

Citation

After using the CombInference R package, please cite both of the following:

	@article{zhang2024startrek,
		title={StarTrek: Combinatorial variable selection with false discovery rate control},
		author={Zhang, Lu and Lu, Junwei},
		journal={The Annals of Statistics},
		volume={52},
		number={1},
		pages={78--102},
		year={2024},
		publisher={Institute of Mathematical Statistics}
	}

---