In clinical applications, single-cell- or bulk RNA sequencing data are commonly used for pre-dicting clinical outcomes, like therapy response. These sequencing data are typically high-dimensional...Show moreIn clinical applications, single-cell- or bulk RNA sequencing data are commonly used for pre-dicting clinical outcomes, like therapy response. These sequencing data are typically high-dimensional where the number of covariates is larger than the sample size. Single-cell RNAsequencing data stand out by carrying more information than the bulk data but are limited bytheir high cost for widespread use. Therefore, practitioners seek to deconvolute bulk data intosingle-cell data for further use to save budgets. Recently, a Bayesian Log-normAl DEconvolution(BLADE) method is developed to tackle this problem. BLADE results in not only the single-cellgene expression profile estimates but also their covariate uncertainties. We, therefore, require amodel to account for these requirements, i.e. generalized linear regression, high dimensionality,and covariate-specific uncertainties.Traditional measurement error models can account for covariate-specific uncertainties, butcannot handle high-dimensionality. Hence, we turn to the Generalized Matrix Uncertainty Selec-tor (GMUS), which accounts for all requirements. GMUS assumes that all measurement errorsare bound by one overall error bound. We extend GMUS to incorporate the covariate uncer-tainties by inserting the covariate measurement error matrix, called Covariate-specific GMUS(CGMUS). We demonstrate CGMUS in simulation studies and an application in cancer ge-nomics. Simulation results show that CGMUS improves the covariate selection precision whenthe covariate measurement errors are small, but fails when some measurement errors are muchlarger than others. Additionally, the CGMUS shows worse prediction performance than GMUSin most simulation settings. The application results show a situation when CGMUS may fail.We conclude that our method is limited to practical use. Several underlying assumptions needto be adapted to improve the performance when covariate uncertainties are incorporated.Show less