Hypoxia classifier for transcriptome datasets.

Area of Science:

• Molecular biology
• Genomics
• Bioinformatics

Background:

• Existing gene signatures for cell states are often limited in scope and applicability.
• Characterizing cellular response to hypoxia requires robust and widely applicable tools.

Purpose of the Study:

• To develop broadly applicable gene classifiers for identifying hypoxia.
• To leverage a large meta-analysis of RNA-sequencing data for classifier development.

Main Methods:

• Conducted a meta-analysis of 69 differential expression datasets (425 RNA-seq experiments) across 33 human cell types under varying hypoxia conditions.
• Developed decision tree classifiers based on a small set of key hypoxia-responsive genes.
• Validated classifiers using cross-validation and independent challenging datasets.

Main Results:

• Generated simple decision tree classifiers (3-5 genes) with over 95% accuracy in cross-validation and >80% on external datasets.
• Successfully identified hypoxic tumor samples and hypoxic regions within tumors using bulk and spatially resolved transcriptomics.
• Discovered a novel hypoxic gene expression pattern in the kidney cortex.

Conclusions:

• The developed gene classifiers are accurate, broadly applicable, and outperform traditional hypoxia signatures.
• These classifiers offer a simplified yet powerful approach for hypoxia detection in various biological contexts and data types.
• The findings have implications for cancer research and understanding tissue physiology.