Variable Selection Techniques in High-Dimensional Genomic Data

The rapid emergence of high-throughput genomic profiling has led to datasets encompassing tens of thousands to millions of measurements per sample. In such settings, classical statistical techniques struggle when the number of variables far exceeds the number of observations. Variable selection techniques address this imbalance by identifying a sparse subset of relevant genetic features while controlling for noise and correlation among markers. Methods based on penalised regression, including the least absolute shrinkage and selection operator (Lasso) and its extensions, introduce regularisation penalties to impose sparsity and enhance interpretability. Group-wise penalties such as group Lasso and overlapping group screening incorporate prior biological knowledge by selecting clusters of genes or pathways, thereby capturing gene–gene and gene–environment interactions. Bayesian frameworks adopt spike-and-slab priors or other hierarchical models to quantify uncertainty and to accommodate complex dependence structures. Screening methods perform an initial dimension reduction by ranking marginal associations before applying more sophisticated selection. Collectively, these approaches facilitate biomarker discovery, improve predictive modelling for disease risk and treatment response, and uncover biological mechanisms across diverse populations.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Variable Selection Techniques in High-Dimensional Genomic Data publication trend

The graph below shows the total number of articles in variable selection techniques in high-dimensional genomic data across all publications each year (not limited to Nature Index journals).

Technical terms

High-dimensional data: Datasets in which the number of variables far exceeds the number of observations, common in genomic studies.

Lasso: A penalised regression method that imposes an l₁-norm penalty to induce sparsity in coefficient estimates.

Group Lasso: An extension of Lasso that applies penalties to predefined groups of variables, promoting the selection of entire pathways or gene sets.

Overlapping group screening: A two-step procedure that first screens variables by group overlap using prior annotations, then applies regularisation to select relevant features and interactions.

Spike-and-slab prior: A Bayesian prior combining a point mass at zero with a diffuse component, enabling variable selection by distinguishing between active and inactive effects.

References

1. Overlapping group screening for detection of gene-gene interactions: application to gene expression profiles with survival trait. BMC Bioinformatics (2018).
2. Overlapping group screening for detection of gene-environment interactions with application to TCGA high-dimensional survival genomic data. BMC Bioinformatics (2022).
3. Lasso estimation of hierarchical interactions for analyzing heterogeneity of treatment effect. Statistics in Medicine (2021).
4. Identifying Gene–Environment Interactions With Robust Marginal Bayesian Variable Selection. Frontiers in Genetics (2021).