Semi-supervised deep matrix factorization model for clustering multi-omics data.

Area of Science:

• Computational Biology
• Bioinformatics
• Data Science

Background:

• Multi-omics data present challenges due to high dimensionality, sparsity, and noise.
• Conventional methods struggle with noise, interpretability, and capturing non-linear patterns.
• Existing multi-view non-negative matrix factorization methods are largely unsupervised.

Purpose of the Study:

• To develop a robust model for multi-omics data integration and clustering.
• To address limitations of existing methods in handling complex, high-dimensional biological data.
• To leverage both labeled and unlabeled samples for improved performance.

Main Methods:

• Proposed SSD-MO (Semi-Supervised Deep Non-Negative Matrix Factorization) model.
• Combines semi-supervised learning with a deep factorization framework.
• Incorporates constraints for preserving geometric structure, orthogonality, and diversity.

Main Results:

• SSD-MO significantly improved clustering accuracy across six multi-omics datasets.
• Achieved 9%-24% increase in F-score over unsupervised baselines.
• Demonstrated robust performance with Precision (64%-73%) and Recall (70%-88%).

Conclusions:

• SSD-MO offers a robust framework for multi-omics data integration.
• The method shows promise for applications in genomics and precision medicine.
• Enhances clustering performance by effectively utilizing labeled and unlabeled data.