Gene expression data classification using topology and machine learning models.

Area of Science:

• Computational Biology
• Genomics
• Data Science

Background:

• Interpreting high-throughput gene expression data requires advanced mathematical tools for high-dimensional analysis.
• Topological Data Analysis (TDA) offers robust feature extraction for complex datasets.
• Existing TDA methods often use 'barcodes' which lack direct data mapping, limiting interpretability.

Purpose of the Study:

• To develop novel TDA-based methods for curating gene expression data.
• To improve feature representation for enhanced biological interpretation and classification.
• To identify representative topological features that directly relate to phenotype labels.

Main Methods:

• Utilized recent developments in TDA to curate gene expression datasets.
• Focused on generating representative persistent cycles instead of traditional barcodes.
• Applied TDA-curated features to supervised classification models (shallow and deep learning).

Main Results:

• The topology-curated data significantly improved classification accuracy in both shallow and deep learning models.
• Computed representative cycles demonstrated an unsupervised correlation with phenotype labels.
• Validated the ability of topological signatures to comprehend gene expression patterns for cohort classification.

Conclusions:

• Generated representative persistent cycles effectively discern gene expression data.
• These cycles provide a direct method for identifying genes involved in similar biological processes.
• TDA offers a powerful framework for analyzing and interpreting complex gene expression datasets.