Multi-view data (collected on the same subjects from different sources) is increasingly common in the biomedical world. The Cancer Genome Atlas project alone has concurrent gene expression, methylation, metabolomics, etc. Traditional methods perform separate analyses on each source, however joint analysis can lead to improved inference and predictions. One of the key challenges for joint analysis is a mixed type of measurements across views (e.g. continuous, binary, ordinal, zero-inflated). Accurate estimation of correlations is often the first critical step in statistical analysis workflows, however Pearson correlation is not well suited for mixed data types as the underlying normality assumption is violated. In this talk, I will demonstrate how latent correlations from Gaussian copula framework provide an elegant alternative to Pearson correlations and a unified approach for treatment of mixed data types. I will illustrate the application of the framework for the analysis of associations between gene expression and microRNA data of breast cancer patients, and for inferring the conditional independence graph in quantitate gut microbiome data.

More information on the speaker: https://irinagain.github.io/bio/