NetMix

NetMix estimates altered subnetworks in biological interaction networks using a network-structured Gaussian Mixture Model to provide less biased identification of differentially expressed genes and cancer driver genes from microarray, RNA-seq, and somatic mutation datasets.

Key Features:

• Reduced Bias Estimation: Employs a Gaussian Mixture Model (GMM) approach to reduce bias in estimates of altered subnetwork sizes that can be inflated by existing methods.
• Statistical Rigor: Formulates altered-subnetwork detection as parameter estimation of an Altered Subset Distribution (ASD) and connects methods such as jActiveModules to their maximum likelihood estimators to reveal statistical biases.
• Improved Performance: Demonstrates superior performance on simulated and real datasets, including detection of differentially expressed genes from microarray and RNA-seq data and identification of cancer driver genes from somatic mutation datasets.

Scientific Applications:

• Differential expression analysis: Identifies altered subnetworks corresponding to differentially expressed genes from microarray and RNA-seq experiments.
• Cancer driver gene identification: Pinpoints candidate cancer driver genes within somatic mutation datasets by locating altered subnetworks.
• Network subnetwork delineation: Enhances interpretability of biological interaction networks through more precise subnetwork delineation.

Methodology:

Uses a network-structured mixture model employing Gaussian Mixture Models (GMM), formulates parameter estimation under the Altered Subset Distribution (ASD), relates jActiveModules to maximum likelihood estimators, and evaluates performance on simulated and real datasets.