CSAX: Characterizing Systematic Anomalies in eXpression Data.

Area of Science:

• Computational biology
• Genomics
• Machine learning

Background:

• Traditional gene expression analysis requires numerous similar patient samples.
• Detecting rare or unique disease abnormalities is computationally challenging.
• Existing methods struggle with high-dimensional and variable expression data.

Purpose of the Study:

• To develop a novel computational approach for anomaly detection in gene expression data.
• To identify disrupted biological pathways within individual patient samples.
• To facilitate the identification of unique disease patterns for precision medicine.

Main Methods:

• Developed a novel computational approach named CSAX (Computational Signature Anomaly Explorer).
• Compiled and released a compendium of public expression datasets for anomaly detection testing.
• Evaluated CSAX's accuracy and compared it to leading anomaly detection methods.

Main Results:

• CSAX accurately identifies anomalies and explains their underlying biological context.
• Demonstrated CSAX's effectiveness on a diverse compendium of expression datasets.
• Successfully applied CSAX to identify pathway disruptions in specific developmental cases.

Conclusions:

• CSAX offers a powerful tool for identifying individual disease patterns from gene expression data.
• The method highlights disrupted platelet activation in retinopathy of prematurity.
• CSAX identified dysregulated oxidative stress in fetuses of obese mothers, a novel finding.