Identifying communities from multiplex biological networks by randomized optimization of modularity

Area of Science:

• Computational Biology
• Network Analysis
• Bioinformatics

Background:

• Community detection is crucial for analyzing large biological networks.
• The Disease Module Identification (DMI) DREAM challenge benchmarked clustering methods for biomedical relevance.
• Evaluating community associations with GWAS-derived genes is key for disease gene discovery.

Purpose of the Study:

• To implement extensions to the MolTi software for improved community detection in biological networks.
• To evaluate the performance of these enhanced MolTi features on simulated and benchmark datasets.
• To assess the impact of randomization, edge weights, and layer weights on detecting trait and disease communities.

Main Methods:

• Extended MolTi software with a randomized Louvain algorithm, edge/layer weighting, and recursive clustering.
• Tested performance on simulated networks to validate community detection improvements.
• Applied enhanced MolTi to the DMI DREAM challenge benchmark datasets.

Main Results:

• Randomized Louvain algorithm significantly improved community detection on simulated networks.
• Performance on the DMI benchmark varied with Genome-Wide Association Studies (GWAS) datasets and enrichment thresholds.
• Weighted edges/layers and randomization generally increased the number of detected trait and disease communities.

Conclusions:

• The enhanced MolTi software offers improved capabilities for biological network community detection.
• The choice of GWAS dataset and statistical threshold critically influences results in disease module identification.
• The implemented features provide a more robust approach to identifying disease-associated modules in biological networks.