Iterative single-cell multi-omic integration using online learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Integrating large single-cell datasets (gene expression, chromatin accessibility, DNA methylation) demands scalable computational tools.
• Existing methods struggle with the increasing volume and continuous arrival of single-cell data.

Purpose of the Study:

• To introduce online integrative non-negative matrix factorization (iNMF) for efficient and scalable integration of large, diverse, and streaming single-cell datasets.
• To enable simultaneous analysis of large datasets by multiple users via internet streaming.
• To facilitate the mapping of new data to existing reference datasets.

Main Methods:

• Developed an online integrative non-negative matrix factorization (iNMF) algorithm.
• Implemented fixed-memory scaling for arbitrarily large cell numbers.
• Enabled iterative incorporation of new datasets.
• Facilitated internet-based data streaming for multi-user analysis.

Main Results:

• Online iNMF effectively integrates over 1 million cells on a standard laptop.
• The method successfully integrates single-cell RNA sequencing and spatial transcriptomic data.
• Demonstrated iterative construction of a multi-omic atlas for the mouse motor cortex.
• Showed no sacrifice in dataset alignment or cluster preservation compared to previous methods.

Conclusions:

• Online iNMF provides a general, scalable, and efficient solution for integrating large and diverse single-cell omics data.
• The algorithm supports continuous data streams and multi-user analysis, advancing the creation of comprehensive biological atlases.
• This approach enhances the feasibility of large-scale multi-omic data integration and analysis.