MONET : a toolbox integrating top-performing methods for network modularisation

Mattia Tomasoni, Sergio Gómez, Jake Crawford, Weijia Zhang, Sarvenaz Choobdar, Daniel Marbach, Sven BergmannbioRxiv 2019 · DOI

Abstract

We define a disease module as a partition of a molecular network whose components are jointly associated with one or several diseases or risk factors thereof. Identification of such modules, across different types of networks, has great potential for elucidating disease mechanisms and establishing new powerful biomarkers. To this end, we launched the “Disease Module Identification (DMI) DREAM Challenge”, a community effort to build and evaluate unsupervised molecular network modularisation algorithms. Here we present MONET, a toolbox providing easy and unified access to the three top methods from the DMI DREAM Challenge for the bioinformatics community. MONET is a command line tool for Linux, based on Docker and Singularity containers; the core algorithms were written in R, Python, Ada and C++.

Introduction

In modern biology, genes and proteins rarely function in isolation. Instead, they operate within intricate molecular networks, such as protein-protein interaction maps or signaling pathways. A "disease module" refers to a cluster of genes within these networks that are collectively associated with a specific pathology. Identifying these modules is a cornerstone of precision medicine, offering pathways to discover novel biomarkers and therapeutic targets.

However, researchers face two primary challenges: algorithmic uncertainty and software usability. With dozens of clustering methods available, identifying the most effective approach was difficult until the "Disease Module Identification (DMI) DREAM Challenge" systematically evaluated 75 different methods. Furthermore, even the top-performing algorithms often suffer from "dependency hell," being written in disparate languages like Ada or C++ with conflicting library requirements. MONET (MOlecular Network Analysis Tool) addresses these issues by integrating the challenge winners into a single, containerized environment.

The MONET Framework

MONET provides a unified command-line interface for the three top-performing algorithms from the DREAM Challenge. By utilizing Docker and Singularity virtualization, MONET ensures that these tools run in isolated "sandboxes" with all necessary dependencies pre-installed. This approach guarantees reproducibility across different Linux environments without the need for manual configuration.

Integrated Methods

  1. K1 (Kernel Clustering Optimization): The overall challenge winner. It utilizes "Diffusion State Distance" to identify functional similarities between genes, followed by spectral clustering to define modules.
  2. M1 (Modularity Optimization): A multiresolution approach that employs a "resistance" parameter to detect modules at various topological scales using an ensemble of optimization techniques like Tabu search.
  3. R1 (Random Walk): Based on a balanced Markov Cluster Algorithm (bMLRMCL), this method is optimized for massive graphs, preventing the formation of oversized, biologically uninformative clusters.

Performance and Scalability

The performance of these methods was rigorously tested using both simulated LFR benchmarks and real-world Genome-wide Association Study (GWAS) data. A key finding is the inherent trade-off between accuracy and computational cost.

Clustering performance (NMI) of the three MONET methods compared to the Louvain algorithm across varying mixing parameters.

As shown in the performance analysis, K1 and M1 maintain high Normalized Mutual Information (NMI) scores even as the mixing parameter (the fraction of inter-community edges) increases, significantly outperforming standard methods like Louvain. However, this accuracy comes at a cost of computational resources.

Maximum RAM usage as a function of network size, highlighting the extreme resource efficiency of the R1 and M1 methods compared to the heavyweight K1 algorithm.

Resource monitoring reveals that while K1 is the "heavyweight" in terms of memory and runtime, R1 acts as a "sprinter," capable of processing massive datasets in under an hour with minimal RAM usage. M1 serves as a versatile "all-rounder," balancing speed and accuracy effectively.

Discussion and Conclusion

MONET demonstrates that no single modularisation method is universally superior; rather, the methods are often complementary. By providing access to all three, MONET allows researchers to gain a more comprehensive biological perspective. While the tool currently requires some manual parameter tuning for peak performance (particularly for R1) and is limited to static network analysis on Linux, it represents a significant step forward in making high-performance bioinformatics tools accessible and reproducible. MONET is freely available at https://github.com/BergmannLab/MONET.git.

Summary updated

Abstract

Abstract Summary We define a disease module as a partition of a molecular network whose components are jointly associated with one or several diseases or risk factors thereof. Identification of such modules, across different types of networks, has great potential for elucidating disease mechanisms and establishing new powerful bio-markers. To this end, we launched the “Disease Module Identification (DMI) DREAM Challenge”, a community effort to build and evaluate unsupervised molecular network modularisation algorithms (Choobdar et al ., 2018). Here we present MONET , a toolbox providing easy and unified access to the three top methods from the DMI DREAM Challenge for the bioinformatics community. Availability and Implementation MONET is a command line tool for Linux, based on Docker and Singularity containers; the core algorithms were written in R, Python, Ada and C++. It is freely available for download at https://github.com/BergmannLab/MONET.git Contact [email protected] (MT); [email protected] (SB)
BibTeX
@misc{tomasoni2019i,
  author = {Mattia Tomasoni and Sergio Gómez and Jake Crawford and Weijia Zhang and Sarvenaz Choobdar and Daniel Marbach and Sven Bergmann},
  title = {{<i>MONET</i> : a toolbox integrating top-performing methods for network modularisation}},
  year = {2019},
  doi = {10.1101/611418},
  howpublished = {bioRxiv},
}
Figure