Gemma - Multi-Omics Toolbox APPROACH

The toolbox will support multiple integration strategies depending on the research question and data availability. Late integration is implemented by processing each omics layer independently (e.g., calling variants, detecting differentially methylated regions, profiling microbial differences) and combining the results at the interpretation stage, through pathway enrichment, network analyses, or meta-analyses. This approach is straightforward, robust to technical differences across datasets, and provides biologically interpretable outcomes.

Another option for integration is mid integration, which is achieved by feeding the different omics layers into a joint statistical or machine learning model , which identifies shared latent factors across data types while preserving their specific structures. This approach is more powerful for uncovering cross-omics interactions and mechanistic links.

With both late and mid integration modules, the toolbox ensures flexibility: researchers can choose simpler workflows when appropriate, but also apply state-of-the-art integrative modeling for comprehensive analyses.

Last option in the toolbox is a graph-based integration model. It is an integration strategy based on logical connections among nodes that represent omics data. Reactome Pathways database is then used to glue different omics domains according to their participation in biological reactions.

In GEMMA project the mid integration strategy for is used for integrating microbiome, methylation and genome data, and further late integration strategy for associating metabolomics, immunoprofiling and proteome data with the other omics measurements.