Bayesian subset-based gene selection for biomarker discovery in high-dimensional data

We introduce the Bayesian Subset-based Gene Selection (BSGS) algorithm, a novel framework tailored for high-dimensional gene expression analysis. BSGS integrates Lasso-based preselection, stochastic subset sampling, and an iterative Bayesian update mechanism to identify a parsimonious yet highly informative set of genes. Central to its methodology is a weighted likelihood function that fuses each gene’s individual contribution with the collective “group score”, which synthesizes key statistical measures—such as a normalized Signal-to-Noise Ratio (Norm-SNR), correlation, variance, and normalized Lasso coefficients—into a unified scoring metric. Through rigorous evaluation on real-world transcriptomics datasets, BSGS consistently demonstrates strong performance in biomarker discovery, striking an optimal balance between interpretability, statistical robustness, and biological relevance. These findings highlight the potential of BSGS as a powerful tool in bioinformatics and precision medicine by facilitating the identification of critical biomarkers for disease diagnosis and treatment.