latent Gaussian copula model

latent Gaussian copula model models dependence in mixed continuous and discrete datasets to integrate fMRI and single nucleotide polymorphism (SNP) data and infer association networks in imaging genetics.

Key Features:

• Integration of Mixed Data Types: Handles continuous, binary, count, and multinomial data to integrate fMRI and SNP datasets.
• Latent Variable Approach: Represents discrete variables, such as SNPs, as discretized versions of underlying continuous latent variables.
• Graphical Model Structure: Employs a graphical framework to represent interactions among variables and uses a semi-rank based estimator to infer the graph structure.
• Robust Performance: Simulation studies report more stable and accurate recovery of underlying graph structures for mixed data compared to existing methods.

Scientific Applications:

• Brain Imaging Genetics: Reveals associations between genetic variations (SNPs) and brain activity patterns measured by fMRI to explore neurogenetic mechanisms.
• Schizophrenia Research: Applied to Mind Clinical Imaging Consortium (MCIC) datasets focused on schizophrenia to identify biologically relevant SNP–brain associations.

Methodology:

Assumes a latent Gaussian structure for continuous variables, models discrete/multinomial data by discretizing latent continuous variables, and uses a semi-rank based estimator to construct the graphical model and identify associations.