Cancer outlier analysis based on mixture modeling of gene expression data

Area of Science:

• Oncology
• Bioinformatics
• Statistical Genetics

Background:

• Cancer molecular heterogeneity arises from genetic alterations, leading to diverse gene expression profiles.
• Detecting cancer-related genes active in subsets of samples (outliers) is challenging due to low statistical power compared to standard methods.
• Existing outlier detection methods often lack power when analyzing complex gene expression data.

Purpose of the Study:

• To develop a novel statistical method for identifying cancer outlier genes with improved power.
• To leverage information sharing across genes and samples for more robust outlier detection.
• To apply the method to real-world cancer datasets for validation.

Main Methods:

• Utilized parametric normal mixture modeling for gene expression levels across cancer samples.
• Standardized gene expression data using reference normal samples.
• Developed a gene-based statistic using posterior probability of cancer outlier for each sample.

Main Results:

• Demonstrated efficiency improvements in detecting cancer outlier genes compared to existing methods.
• The proposed method showed robustness even with misspecified heavy-tailed t-distributions.
• Successfully applied the method to a real dataset of hematologic malignancies.

Conclusions:

• The developed method enhances the power of cancer outlier gene detection by integrating information across samples and genes.
• Parametric normal mixture modeling provides a flexible framework for analyzing complex cancer gene expression data.
• This approach offers a valuable tool for discovering novel cancer-related genes in heterogeneous tumor samples.