The spike-and-slab quantile LASSO for robust variable selection in cancer genomics studies

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Cancer genomics studies often exhibit data irregularities like outliers and heavy-tailed distributions.
• Robust variable selection methods are crucial for identifying genes linked to heterogeneous disease traits and building accurate predictive models.

Purpose of the Study:

• To develop a robust variable selection method that combines the strengths of quantile LASSO and Bayesian regularized quantile regression for high-dimensional genomics data.
• To overcome limitations of existing methods by proposing a fully Bayesian spike-and-slab formulation using the asymmetric Laplace distribution (ALD).

Main Methods:

• The proposed method, spike-and-slab quantile LASSO, utilizes a robust likelihood based on the asymmetric Laplace distribution (ALD).
• It incorporates selective shrinkage and self-adaptivity properties from spike-and-slab LASSO.
• Computational efficiency is achieved through Expectation-Maximization (EM) steps within a coordinate descent framework.

Main Results:

• Comprehensive simulations demonstrate the superiority of the spike-and-slab quantile LASSO over competing methods, especially with heavy-tailed errors in various settings.
• The method effectively handles both homogeneous and heterogeneous model conditions.

Conclusions:

• The spike-and-slab quantile LASSO offers a robust and computationally advantageous approach for variable selection in cancer genomics.
• Its effectiveness is validated through simulations and applied successfully to lung adenocarcinoma (LUAD) and skin cutaneous melanoma (SKCM) datasets from The Cancer Genome Atlas (TCGA).