Answers are grounded in this paper and cite their sources.
Identifying important variables associated with continuous, binary, or ordered categorical disease-related responses is a key challenge in omics, biosensing, proteomics, and biomedical engineering. Although numerous variable selection methods have been developed for parametric and nonparametric models, many do not adequately account for correlations among predictors. Our motivating data arise from surface-enhanced Raman spectroscopy (SERS), used to investigate ordered drug dosage levels in brain cancer cells. SERS measures Raman signal intensities across multiple wavenumbers and spatial locations within each cell, producing spectral profiles that serve as molecular fingerprints. Identifying informative wavenumbers that distinguish drug dosage levels is particularly challenging due to the correlated structure of these measurements. To address this challenge, we propose a Bayesian variable selection method based on a generalized fused multi-kernel machine regression framework. The proposed approach accommodates continuous, binary, and ordered categorical responses and incorporates variable selection within a Bayesian hierarchical model for nonparametric regression. This framework enables the simultaneous identification of informative predictors while accounting for their correlation structure. We demonstrate the proposed method using SERS molecular profiling data to investigate drug dose–response relationships in brain cancer cells.
No takes yet. Share an insight, caveat, or question.
Jin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: