Multi-omics Data Integration Model Based on UMAP Embedding and Convolutional Neural Network.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration offers deeper insights than single-omic approaches.
• Existing methods for multi-omics analysis are crucial for biomedical applications like disease prediction.

Purpose of the Study:

• To introduce a novel multi-omics data integration method using UMAP and CNNs.
• To enhance disease prediction accuracy by integrating gene expression, DNA methylation, and copy number alteration data.

Main Methods:

• Utilized Uniform Manifold Approximation and Projection (UMAP) to create 2D embeddings from multi-omics data.
• Constructed a gene similarity network (GSN) using gene expression as a reference.
• Employed Convolutional Neural Networks (CNNs) for predicting prostate cancer Gleason scores and breast cancer tumor stages.

Main Results:

• Achieved prediction accuracy exceeding 99% for both prostate cancer Gleason score and breast cancer tumor stage.
• The proposed UMAP-based GSN integration outperformed the state-of-the-art iSOM-GSN model.
• Demonstrated superior integration capabilities of UMAP compared to Self-Organizing Maps (SOM).

Conclusions:

• UMAP is an effective technique for integrating multi-omics data into predictive models.
• The developed model shows potential for predicting various cancer types using multi-omics data.