StableMate: a statistical method to select stable predictors in omics data.

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Statistical associations are key to understanding molecular mechanisms.
• Omics data complexity and variability limit reproducibility and interpretability of traditional association studies.
• Translating associations into robust biological hypotheses remains challenging.

Purpose of the Study:

• To develop a novel regression framework, StableMate, for robust variable selection across heterogeneous omics datasets.
• To address challenges in reproducibility and interpretability of biological associations.
• To identify environment-agnostic (stable) and environment-specific predictors for generalizable biological insights.

Main Methods:

• StableMate employs a variable selection process across heterogeneous datasets.
• It distinguishes between environment-agnostic (stable) and environment-specific predictors.
• The framework is adaptable to both regression and classification analyses for various omics data types.

Main Results:

• StableMate identified stable predictors representing robust functional dependencies, enabling generalizable predictions.
• Applied to breast cancer RNA sequencing data, it discovered genes consistently predicting estrogen receptor expression.
• In metagenomics and single-cell RNA sequencing data, it identified persistent microbial signatures for colon cancer and signature genes for pro-tumour microglia in glioblastoma, respectively.

Conclusions:

• StableMate provides a robust framework for identifying reproducible biological associations from complex omics data.
• The method enhances the interpretability and generalizability of findings across different biological contexts and data types.
• StableMate facilitates comprehensive characterization of biological systems for advanced omics research.