scMNMF: a novel method for single-cell multi-omics clustering based on matrix factorization.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell multi-omics data analysis offers comprehensive cellular insights but faces challenges in clustering due to high dimensionality and sparsity.
• Existing analysis algorithms often exhibit suboptimal clustering performance for complex single-cell multi-omics datasets.

Purpose of the Study:

• To introduce scMNMF, a novel non-negative matrix factorization algorithm for joint dimensionality reduction and cell clustering of single-cell multi-omics data.
• To enhance the discovery of cell types by leveraging iterative feature selection and dimensionality reduction.

Main Methods:

• Developed scMNMF, an algorithm employing non-negative matrix factorization for joint dimensionality reduction and cell clustering.
• Formulated the objective function as a constrained optimization problem with iterative formulas derived via alternating iterative algorithms.
• Implemented a method where feature selection for dimensionality reduction and cell clustering mutually influence each other.

Main Results:

• scMNMF effectively explores hidden related features among different omics data.
• The iterative mutual influence between feature selection and clustering leads to more effective cell type discovery.
• Validated on two simulated and five real datasets, scMNMF demonstrated superior performance compared to seven other state-of-the-art algorithms.

Conclusions:

• scMNMF provides a robust framework for analyzing single-cell multi-omics data, overcoming limitations in clustering performance.
• The algorithm's ability to integrate dimensionality reduction and clustering offers improved cell type identification.
• scMNMF represents a significant advancement in computational tools for single-cell multi-omics research.