A Customizable Analysis Flow in Integrative Multi-Omics

Abstract

The number of researchers using multi-omics is growing. Though still expensive, every year it is cheaper to perform multi-omic studies, often exponentially so. In addition to its increasing accessibility, multi-omics reveals a view of systems biology to an unprecedented depth. Thus, multi-omics can be used to answer a broad range of biological questions in finer resolution than previous methods. We used six omic measurements-four nucleic acid (i.e., genomic, epigenomic, transcriptomics, and metagenomic) and two mass spectrometry (proteomics and metabolomics) based-to highlight an analysis workflow on this type of data, which is often vast. This workflow is not exhaustive of all the omic measurements or analysis methods, but it will provide an experienced or even a novice multi-omic researcher with the tools necessary to analyze their data. This review begins with analyzing a single ome and study design, and then synthesizes best practices in data integration techniques that include machine learning. Furthermore, we delineate methods to validate findings from multi-omic integration. Ultimately, multi-omic integration offers a window into the complexity of molecular interactions and a comprehensive view of systems biology.

Keywords: analysis flow; bioinformatics; machine learning; multi-omics; multi-omics analysis; study design.

Figure 1

The molecules profiled in multi-omics studies. We describe 6 levels of information, starting from the bottom to the top: genome, epigenome, transcriptome, proteome, metabolome, and metagenome. The genome, epigenome, transcriptome, and metagenome are profiled by sequencing-based technologies such as sequencing by synthesis, depicted here, to profile a comprehensive set of nucleic acid molecules. On the other hand, mass spectrometers generate proteome and metabolome profiles as depicted here through measurements of biomolecules’ masses and charges. For overlapping technologies, each omic level provides unique information and insights into cellular activity present in conditions being studied. By leveraging the layers of information, longitudinal and cross-sectional multi-omics studies find modules (e.g., cell signaling pathways) that are differential between healthy and disease states. These modules represent complex system biology networks that give precise insights into the molecular dysregulation in disease states.

Figure 2

Typical multiomic study designs. Gray dots represent samples taken. (a) A case control observational study. A population is taken with participants that have the phenotype of interest (cases) and those without (controls). Cases and controls are sampled in even amounts. (b) A randomized longitudinal study where n participants are randomized into two arms of a study. In this case an increasing treatment dose is administered, and samples are taken every week.