Gene set optimization for cancer transcriptomics using sparse principal component analysis.

Area of Science:

• Bioinformatics
• Cancer Genomics
• Computational Biology

Background:

• Gene set analysis of tumor transcriptomic data is common for exploring cancer pathway dysregulation.
• Existing gene set collections often model normal tissue gene activity, limiting their effectiveness in cancer research.
• Tumor gene activity patterns can significantly differ from normal tissue patterns.

Purpose of the Study:

• To develop a bioinformatics approach for optimizing gene set collections to better represent cancer gene activity.
• To adapt the Molecular Signatures Database (MSigDB) Hallmark collection for 21 solid human cancers using The Tumor Genome Atlas (TCGA) data.
• To improve the biological utility of gene set analysis in cancer research.

Main Methods:

• Developed a bioinformatics approach utilizing sparse principal component analysis (PCA).
• Applied PCA to optimize the MSigDB Hallmark gene set collection.
• Utilized bulk RNA-sequencing data from TCGA for 21 solid human cancers.

Main Results:

• Optimized gene set collections reflect gene activity patterns in dysplastic (cancerous) tissue.
• The average survival association of gene set members improved after optimization.
• This improvement was observed across nearly all cancer types and Hallmark gene sets analyzed.

Conclusions:

• The developed sparse PCA-based bioinformatics approach effectively optimizes gene set collections for cancer research.
• Optimized gene sets enhance the accuracy of pathway dysregulation analysis in tumors.
• The method demonstrates biological utility by improving survival association, offering a valuable tool for cancer genomics and precision medicine.