MOTIVATION: The heterogeneity of complex diseases including cancer leads to heavy-tailed distributions in the disease traits. In such settings, non-robust variable selection methods are inherently susceptible to data contamination and can yield unstable or misleading results. This vulnerability becomes more severe for recently proposed approaches that introduce pseudo-features as negative controls, as these methods further amplify the curse of dimensionality by expanding the genotype matrix in the presence of outliers and high-dimensional genomic features. RESULTS: We develop a robust variable selection framework with stability selection to prioritize genomic features in the presence of contamination. In contrast to existing approaches that rely on pseudo-features for error control, the proposed method achieves double robustness. First, it adopts least absolute deviation (LAD) LASSO to ensure robustness against outliers and heavy-tailed errors in disease traits. Second, it avoids augmenting the genotype matrix with pseudo-features, thereby mitigating the curse of dimensionality that is particularly problematic in high-dimensional genomic data. The proposed method has been extensively evaluated in simulation studies to demonstrate its effectiveness over multiple competing methods for variable selection. In addition, we have applied the proposed method and competing approaches to two real-data case studies: the The Cancer Genome Atlas (TCGA) Skin Cutaneous Melanoma (SKCM) dataset and an eQTL dataset. The results demonstrate that the proposed method achieves superior performance by identifying genomic features with higher reproducibility. AVAILABILITY AND IMPLEMENTATION: The source code for implementing the proposed methods is publicly available at https://github.com/cenwu/RSS with an archival DOI https://doi.org/10.6084/m9.figshare.32306883.

A drastic amount of data have been and are being generated in bioinformatics studies. In the analysis of such data, the standard modeling approaches can be challenged by the heavy-tailed errors and outliers in response variables, the contamination in predictors (which may be caused by, for instance, technical problems in microarray gene expression studies), model mis-specification and others. Robust methods are needed to tackle these challenges. When there are a large number of predictors, variable selection can be as important as estimation. As a generic variable selection and regularization tool, penalization has been extensively adopted. In this article, we provide a selective review of robust penalized variable selection approaches especially designed for high-dimensional data from bioinformatics and biomedical studies. We discuss the robust loss functions, penalty functions and computational algorithms. The theoretical properties and implementation are also briefly examined. Application examples of the robust penalization approaches in representative bioinformatics and biomedical studies are also illustrated.

Set-based association studies based on genes or pathways have shown great promise in interpreting association signals associated with complex diseases. These approaches are particularly useful when variants in a set have moderate effects and are difficult to be detected with single marker analysis, especially when variants function jointly in a complicated manner. The set-based analyses use a summary statistic such as the maximum or average of individual signal (e.g. a chi-square statistic) over all variants in a set, or consider their joint distribution to assess the significance of the set. The signal obtained with this treatment, however, could be potentially diluted when noisy variants are not taken good care of, leading to either inflated false negatives or false positives. Thus, the selection of disease informative single-nucleotide polymorphism (diSNPs) plays a crucial role in improving the power of the set-based association study. In this work, we propose an efficient diSNP selection method based on the information theory. We select diSNP variants by considering their relative information contribution to a disease status, which is different from the usual tag SNP selection. The relative merit of pre-selecting diSNPs in a set-based association analysis is demonstrated through extensive simulation studies and real data analysis.