Bayesian Reciprocal LASSO Quantile Regression for High-Dimensional Genomic Data
Author(s): Zainab S Alsaadi
Abstract: This paper proposes a Bayesian Reciprocal LASSO
Quantile Regression (BRLQR) model designed for high-dimensional genomic data.
The method combines the reciprocal LASSO penalty to enforce sparsity with the
asymmetric structure of quantile regression, allowing for robust estimation
across different parts of the conditional distribution. Using a Bayesian
framework, the model applies the Asymmetric Laplace Distribution as the likelihood
and the inverse Laplace prior to induce the reciprocal LASSO penalty. A
comprehensive simulation study under various error distributions (normal,
Laplace, and contaminated normal) demonstrates the superior performance of
BRLQR in terms of prediction accuracy (MAD), true positive rate (TPR), and
false discovery rate (FDR). The application to real genomic data confirms the
model's practical capability in handling noisy, high-dimensional features,
highlighting its effectiveness in variable selection and robust prediction
across quantiles.
Zainab S Alsaadi. Bayesian Reciprocal LASSO Quantile Regression for High-Dimensional Genomic Data. Int J Stat Appl Math 2025;10(6):13-17. DOI: 10.22271/maths.2025.v10.i6a.2052
