Integrative analysis of multiple diverse omics datasets by sparse group multitask regression.

Area of Science:

• Genomics and Bioinformatics
• Complex Trait Genetics
• Biomarker Discovery

Background:

• High-throughput genome-wide assays identify genetic risk factors for complex traits.
• Existing studies face limitations like small sample sizes and low reproducibility.
• Integrating diverse omics datasets offers potential to enhance biomarker identification power and consistency.

Purpose of the Study:

• To propose a novel integrative method, sparse group multitask regression, for combining diverse omics datasets, platforms, and populations.
• To identify genetic risk genes and factors for complex diseases by integrating multi-omics data.
• To overcome the 'small sample, large variables' problem in genetic association studies.

Main Methods:

• Developed a sparse group multitask regression method combining multitask learning with sparse group regularization.
• Treated biomarker identification in each study as a separate task within a multitask learning framework.
• Introduced sparse group lasso and sparse group ridge penalties, with effective algorithms and a significance test for gene identification.

Main Results:

• Simulation studies demonstrated that the sparse group multitask method significantly outperforms conventional meta-analysis.
• Applied to osteoporosis studies, the method identified 7 significant genes, validated in three independent studies.
• Identified SOD2 as a significant gene, consistent with previous findings, and highlighted TREML2, HTR1E, and GLO1 as novel susceptible genes for osteoporosis.

Conclusions:

• The proposed sparse group multitask regression is an effective integrative method for identifying genetic risk factors of complex diseases from diverse omics data.
• This approach enhances the power and consistency of biomarker discovery compared to traditional methods.
• The identified genes, including novel ones like TREML2, HTR1E, and GLO1, provide new insights into osteoporosis susceptibility.