Rank-based Bayesian variable selection for genome-wide transcriptomic analyses.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional omics data analysis requires robust variable selection.
• Unsupervised variable selection is challenging and often lacks reproducibility.
• Current methods may not fully capture biological relevance in complex datasets.

Purpose of the Study:

• To develop a Bayesian variable selection method for unsupervised transcriptomic analysis.
• To enhance reproducibility and robustness in omics data analysis.
• To identify biologically relevant genes for cancer genomics.

Main Methods:

• A novel extension of the Bayesian Mallows model for rank-based variable selection.
• Utilizing data rankings instead of continuous measurements for increased robustness.
• Integrating variable selection within inferential tasks for complete reproducibility.

Main Results:

• The proposed method demonstrates versatility and robustness across various scenarios in simulation studies.
• It outperforms existing methods in high-dimensional settings and diverse data-generating processes.
• Successfully identified key genes involved in ovarian cancer development from RNA-seq data.

Conclusions:

• The Bayesian rank-based approach offers a reproducible and robust solution for unsupervised variable selection in omics.
• It is effective for signature discovery in cancer genomics, aiding biological investigation.
• Uncertainty quantification is crucial for subsequent biological interpretation and validation.