ClusterMI: Detecting High-Order SNP Interactions Based on Clustering and Mutual Information

Area of Science:

• Genetics and Genomics
• Computational Biology
• Statistical Genetics

Background:

• Identifying single nucleotide polymorphism (SNP) interactions is crucial for understanding complex disease heritability in genome-wide association studies (GWAS).
• Existing methods for detecting SNP interactions face computational challenges due to the vast number of potential high-order interactions.

Purpose of the Study:

• To propose a novel two-stage approach, ClusterMI, for efficiently detecting high-order genome-wide SNP interactions.
• To reduce the computational burden associated with identifying complex genetic interactions.

Main Methods:

• ClusterMI employs a two-stage strategy: a screening stage and a search stage.
• The screening stage uses clustering and mutual information to group SNPs, followed by conditional mutual information to identify significant pairwise SNP combinations within clusters.
• The search stage utilizes exhaustive or ant colony optimization search to detect high-order interactions based on identified pairwise combinations.

Main Results:

• ClusterMI significantly reduces computational load by focusing on promising SNP combinations.
• Extensive simulations demonstrate superior performance of ClusterMI compared to existing approaches.
• Application to real case-control datasets confirms ClusterMI's capability in identifying high-order SNP interactions from genome-wide data.

Conclusions:

• ClusterMI offers an effective and computationally efficient solution for detecting high-order SNP interactions in GWAS.
• The method enhances the ability to uncover complex genetic architectures underlying diseases.