[Paper Review] Signal Subgraph Estimation Via Vertex Screening

Shangsi Wang, Cencheng Shen+3 more|arXiv (Cornell University)|Jan 23, 2018

Bioinformatics and Genomic Networks23 references3 citations

TL;DR

This paper proposes a vertex screening method to identify small, informative subgraphs in complex networks by leveraging distance-based correlation to select signal vertices, significantly improving downstream graph classification and regression performance while enhancing interpretability. The approach is theoretically consistent and empirically validated on functional and structural brain graphs from human and murine MRI data.

ABSTRACT

Graph classification and regression have wide applications in a variety of domains. A graph is a complex and high-dimensional object, which poses great challenges to traditional machine learning algorithms. Accurately and efficiently locating a small signal subgraph dependent on the label of interest can dramatically improve the performance of subsequent statistical inference. Moreover, estimating a signal subgraph can aid humans with interpreting these results. We present a vertex screening method to identify the signal subgraph when given multiple graphs and associated labels. The method utilizes distance-based correlation to screen the vertices, and allows the subsequent classification and regression to be performed on a small induced subgraph. We demonstrate that this method is consistent in recovering signal vertices and leads to better classification performance via theory and numerical experiments. We apply the vertex screening algorithm on human and murine graphs derived from functional and structural magnetic resonance images to analyze the site effects and sex differences.

Motivation & Objective

• To address the challenge of identifying small, informative subgraphs in high-dimensional graph data for improved statistical inference.
• To develop a computationally efficient method that screens vertices based on their correlation with labels, focusing on signal subgraphs.
• To enhance the interpretability of graph-based machine learning models by isolating relevant substructures.
• To demonstrate theoretical consistency and empirical effectiveness of the screening method in real-world neuroimaging applications.

Proposed method

• The method uses distance-based correlation between vertex features and graph labels to rank and screen vertices for signal subgraph identification.
• It induces a subgraph from the top-ranked vertices, reducing dimensionality and focusing analysis on the most relevant network regions.
• The screening procedure is designed to be consistent, ensuring asymptotic recovery of true signal vertices under mild regularity conditions.
• The approach is applied to both classification and regression tasks on the reduced subgraph, improving model performance.
• The method is validated through theoretical analysis and numerical experiments on synthetic and real brain network data.

Experimental results

Research questions

• RQ1 Can vertex screening based on distance-based correlation effectively identify signal subgraphs in labeled graphs?
• RQ2 Does the screening method lead to improved classification and regression performance on reduced subgraphs?
• RQ3 Is the vertex screening procedure theoretically consistent in recovering signal vertices?
• RQ4 How does the method perform in detecting biological effects such as sex differences and site effects in brain networks?

Key findings

• The vertex screening method achieves consistent recovery of signal vertices under mild regularity conditions, ensuring theoretical reliability.
• Classification performance improves significantly when using the screened subgraph compared to full graphs, especially in high-dimensional settings.
• The method effectively detects site effects and sex differences in human and murine brain networks derived from MRI data.
• Numerical experiments confirm the method's superiority in both signal detection and downstream prediction accuracy.