Penalized partial least squares for pleiotropy.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics

Background:

• Genome-wide association studies (GWAS) increasingly identify loci with pleiotropic effects, suggesting shared biological mechanisms across diseases.
• Current methods primarily focus on single nucleotide polymorphism (SNP)-level associations, limiting the discovery of higher-level biological insights.
• Understanding cross-phenotype genetic associations is crucial for elucidating common disease underpinnings.

Purpose of the Study:

• To develop a novel gene- and pathway-level analytical approach for multiple independent GWAS datasets.
• To identify common susceptibility variants and biological mechanisms underlying distinct phenotypes.

Main Methods:

• A generalization of sparse group Partial Least Squares (sgPLS) was developed to integrate multiple independent GWAS datasets.
• The method incorporates Lasso penalization to link datasets and analyze grouped variables, named joint-sgPLS.
• The approach enables detection of signals at both the variable (gene) and group (pathway) levels.

Main Results:

• The joint-sgPLS method provides a global, interpretable model suitable for complex data architectures.
• Performance comparisons on simulated data demonstrated that joint-sgPLS can outperform traditional methods.
• Application to real data highlighted common susceptibility variants for breast and thyroid cancers.

Conclusions:

• Joint-sgPLS effectively detects genetic signals, particularly in datasets with a large number of variables.
• The method's formulation, leveraging Lasso penalization, is adaptable to various data structures and group architectures.
• The approach has broad applicability beyond genetic studies to any field with high-dimensional data and a priori group structures.