A New Approach to Large Multiomics Data Integration.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• High-dimensional omics and imaging data pose challenges for feature extraction and data mining.
• Existing nonlinear dimensionality reduction methods like t-SNE and UMAP excel at visualization but struggle with very large datasets.
• Integrating multi-omics data is crucial for a holistic understanding of systems biology.

Purpose of the Study:

• To develop a new approach for extracting, mining, and integrating large multi-omics datasets.
• To overcome the limitations of current algorithms in handling prohibitively large data.

Main Methods:

• Utilized deep learning on subsampled nonlinear dimensionality reduction (t-SNE and UMAP).
• Applied the method to extract features from mass spectrometry imaging and chromosome conformation capture data.
• Demonstrated learning embeddings from fused omics data, projecting metabolomics into a reduced transcriptomics representation.

Main Results:

• Successfully extracted features from large, complex datasets previously considered too large.
• Enabled the fusion of different omics data through learned embeddings.
• Showcased the projection of metabolomics data into a reduced transcriptomics space.

Conclusions:

• The proposed deep learning approach effectively integrates large and multi-omics data.
• This method advances the analysis of complex biological datasets, enabling new insights in systems biology.
• Facilitates a more comprehensive understanding by fusing diverse biological information streams.