Sorting Five Human Tumor Types Reveals Specific Biomarkers and Background Classification Genes.

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• The Cancer Genome Atlas (TCGA) provides a rich dataset for exploring complex biological patterns.
• Understanding gene expression profiles is crucial for cancer subtyping and biomarker discovery.

Purpose of the Study:

• To investigate the utility of knowledge-independent data-mining methods for cancer subtyping using RNA expression data.
• To identify core biomarker transcripts and assess the robustness of classification after their removal.

Main Methods:

• Application of Dynamic Quantum Clustering (DQC) and t-Distributed Stochastic Neighbor Embedding (t-SNE) to TCGA RNA sequencing data.
• Iterative removal of top biomarker transcripts identified by DQC and subsequent cluster analysis.
• Analysis of tumor classification accuracy and biological function patterns at each iteration.

Main Results:

• RNA expression patterns successfully sorted 2,016 samples from five tumor types by clinical annotations.
• DQC identified 48 core biomarker transcripts that initially clustered tumors by type.
• Tumor classification remained robust even after iterative removal of these biomarkers, suggesting a "background classification" potential.
• Repeating patterns of biological function were detected in later iterations, not apparent with core biomarkers present.

Conclusions:

• Gene expression data possesses inherent classification capabilities beyond known biomarkers.
• The identified "background classification" potential offers new avenues for understanding tumor heterogeneity and developing diagnostic tools.
• Dynamic Quantum Clustering and t-SNE are effective tools for uncovering complex patterns in large-scale cancer genomics data.