GMMchi: gene expression clustering using Gaussian mixture modeling

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Cancer evolves through genetic and epigenetic changes, altering gene expression and cell phenotypes.
• Driver mutations confer selective advantages, influencing cancer gene expression and phenotypes.
• Bimodal gene expression patterns may correlate with driver mutations.

Purpose of the Study:

• To introduce GMMchi, a Python package for detecting and characterizing bimodal gene expression patterns in cancer.
• To analyze correlations between bimodal gene expression shifts and driver mutations.
• To provide a tool for enhanced analysis of bulk gene expression data.

Main Methods:

• Utilized Gaussian Mixture Modeling (GMM) to identify bimodal gene expression patterns.
• Employed 2x2 contingency table statistics for correlation analysis.
• Validated GMMchi using simulated data and applied it to microarray and RNA-Seq data.

Main Results:

• GMMchi demonstrated robust performance with 85% accuracy on simulated data (n=90).
• The package can characterize background signals, filter probes, and uncover genetic interrelationships.
• Successfully extracted bimodal patterns from colorectal cancer (CRC) cell line and tumor data.

Conclusions:

• GMMchi reliably detects bimodal gene expression patterns in cancer datasets.
• The tool verifies known gene expression correlates of CRC phenotypes.
• Offers a valuable addition to traditional continuous-valued statistical analysis in cancer research.