Evaluation and Aggregation of Active Module Identification Algorithms.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• High-throughput sequencing generates extensive genetic data for candidate gene studies.
• Gene-Gene Interaction (GGI) networks are crucial for analyzing candidate genes but can become complex.
• Active Module Identification (AMI) is essential for reducing GGI network complexity by finding enriched subnetworks.

Purpose of the Study:

• To comprehensively assess and compare the performance of four AMI algorithms: PAPER, DOMINO, FDRnet, and HotNet2.
• To evaluate the ability of these algorithms to identify context-specific biological enrichments in GGI networks.
• To propose methods for aggregating results from multiple AMI algorithms to enhance biological signal detection.

Main Methods:

• The Empirical Pipeline (EMP) was used to evaluate four AMI algorithms on four distinct biological datasets.
• Performance was assessed based on the algorithms' ability to produce context-specific enrichment.
• Two novel aggregation methods were developed: spectral clustering and Greedy Conductance-based Merging (GCM).

Main Results:

• No single AMI algorithm demonstrated superior performance across all tested biological datasets.
• Outputs from different AMI algorithms were often dissimilar, indicating they capture complementary biological information.
• The proposed aggregation methods (spectral clustering and GCM) can effectively combine outputs from multiple algorithms.

Conclusions:

• A comprehensive analysis of GGI networks using AMI requires integrating results from multiple algorithms.
• The developed tools and workflows enhance the analysis capabilities for researchers using AMI algorithms.
• Freely available code and resources facilitate broader adoption and application of these methods.