Dealing with dimensionality: the application of machine learning to multi-omics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Machine learning (ML) automates information extraction from large datasets.
• Integrative analysis of multi-omics data is crucial for data-driven tasks.
• Next-generation sequencing and other omics assays generate vast amounts of complex data.

Purpose of the Study:

• To systematically assess the literature on ML for multi-omics data integration.
• To identify key trends, methodologies, and applications in the field.
• To explore how ML addresses the challenges of high-dimensional omics datasets.

Main Methods:

• Literature survey and quantitative exploration of ML techniques in multi-omics.
• Analysis of goals, techniques, and datasets used in ML multi-omics integration.
• Focus on methods addressing datasets with few samples and many features.

Main Results:

• Popular ML techniques include autoencoders, random forests, and support vector machines.
• Dimensionality reduction is frequently employed to manage high feature counts.
• The Cancer Genome Atlas dataset is a dominant resource in this research area.

Conclusions:

• ML methods are vital for extracting insights from complex omics data.
• The field prioritizes techniques robust to high-dimensional, low-sample-size data.
• Standardized datasets like TCGA facilitate reproducible research in ML multi-omics.