Multi-scale analysis and clustering of co-expression networks

TL;DR

This paper introduces a multi-scale network clustering approach to analyze co-expression networks from time-course genomic data, revealing cross-stress features and integrating diverse datasets.

Contribution

It presents a novel multi-scale clustering protocol for co-expression networks that captures cross-stress features and handles diverse datasets.

Findings

Effective network partitioning algorithms identified stress-specific modules.

Cross-stress features were successfully extracted from network communities.

The approach allows integration of datasets with different experimental conditions.

Abstract

The increasing capacity of high-throughput genomic technologies for generating time-course data has stimulated a rich debate on the most appropriate methods to highlight crucial aspects of data structure. In this work, we address the problem of sparse co-expression network representation of several time-course stress responses in {\it Saccharomyces cerevisiae}. We quantify the information preserved from the original datasets under a graph-theoretical framework and evaluate how cross-stress features can be identified. This is performed both from a node and a network community organization point of view. Cluster analysis, here viewed as a problem of network partitioning, is achieved under state-of-the-art algorithms relying on the properties of stochastic processes on the constructed graphs. Relative performance with respect to a metric-free Bayesian clustering analysis is evaluated and possible extensions are discussed. We further cluster the stress-induced co-expression networks generated independently by using their community organization at multiple scales. This type of protocol allows for an integration of multiple datasets that may not be immediately comparable, either due to diverse experimental variations or because they represent different types of information about the same genes.