Harmonizing Multi-Omics for Enhanced Machine Learning

Abstract

The proliferation of high-throughput
technologies has yielded an abundance of omics data,
spanning diverse biological layers such as genomics,
epigenomics, transcriptomics, proteomics, and
metabolomics. Machine learning algorithms have
harnessed this data deluge, yielding diagnostic and
classification biomarkers. However, prevailing biomarkers
predominantly rely on single omic measurements,
overlooking the potential insights from multi-omics
experiments that encapsulate the entirety of biological
complexity. To fully exploit the wealth of information
embedded in different omics layers, effective multi-omics
data integration strategies become imperative. This
minireview categorizes recent integration
methods/frameworks into five strategies: early, mixed,
intermediate, late, and hierarchical. Our focus is on
delineating challenges and exploring existing multi-omics
integration strategies, with a keen emphasis on their
application in machine learning