Determination of nonlinear genetic architecture using compressed sensing

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Extracting genetic architecture for complex traits from genotype-phenotype data is a fundamental challenge in genomics.
• Nonlinear gene-gene interactions (epistasis) and a large number of candidate genes complicate genotype-phenotype association studies.
• Compressed Sensing offers a powerful approach to solve under-constrained systems, outperforming traditional single-variant regression.

Purpose of the Study:

• To introduce a novel Compressed Sensing method for reconstructing nonlinear genetic models from genome-wide association study (GWAS) data.
• To enable the identification of complex gene-gene interactions, including epistasis, which are crucial for understanding complex traits.
• To provide a computationally efficient method for genotype-phenotype association analysis.

Main Methods:

• Application of L1-penalized regression on nonlinear functions of the sensing matrix.
• Utilizing Compressed Sensing principles to solve for sparse nonlinear genetic models.
• Demonstration on simulated human genomes and limited real-world data.

Main Results:

• The method's computational and data requirements are comparable to linear model reconstruction.
• The approach effectively reconstructs nonlinear genetic models under a generalized sparsity assumption.
• A phase transition indicates sufficient data availability for successful model reconstruction.

Conclusions:

• Predictive models for complex traits and human disease susceptibilities can be extracted from datasets with hundreds of individuals.
• The method's efficacy is demonstrated for traits with additive heritability (h^2 ~ 0.5).
• Accurate reconstruction is feasible even for traits influenced by thousands of loci, requiring approximately a million individuals.