Multi-omics network inference with a Gaussian copula model.

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Partial correlation network inference is crucial for identifying direct biological interactions.
• Traditional Gaussian graphical models fail with heterogeneous multi-omics data (continuous and discrete variables).

Purpose of the Study:

• To propose a novel, likelihood-based network inference method for multi-omics data.
• To address limitations of existing methods with discrete and heterogeneous data types.

Main Methods:

• Utilized a Gaussian copula model with semiparametric pairwise-likelihood estimation.
• Employed graphical lasso for inverting and regularizing the latent correlation matrix.
• Recovered latent partial correlations for network construction.

Main Results:

• The proposed method shows improved computational efficiency and estimation accuracy compared to moment-based approaches.
• Demonstrated superior performance with discrete data, including count data (e.g., RNA-seq).
• Successfully identified biologically relevant interactions in a breast cancer dataset (ICGC).

Conclusions:

• The Gaussian copula and likelihood-based approach offers an effective framework for multi-omics network inference.
• The method is computationally efficient and suitable for integrative data analysis.
• Provides a robust tool for uncovering biological interactions from complex datasets.