Multisource single-cell data integration by MAW barycenter for Gaussian mixture models.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Integrating single-cell data from multiple sources presents a significant clustering challenge.
• Existing methods struggle with combining diverse datasets effectively.
• Gaussian mixture models (GMMs) are used to represent individual clustering results.

Purpose of the Study:

• To develop a computationally feasible and accurate method for integrating clustering results from multiple single-cell datasets.
• To address the limitations of traditional Wasserstein barycenter calculations for GMMs.
• To improve the performance of single-cell data analysis through enhanced clustering integration.

Main Methods:

• Representing each dataset's clustering as a Gaussian mixture model (GMM).
• Employing the minimized aggregated Wasserstein (MAW) distance to approximate the Wasserstein metric between GMMs.
• Developing a new algorithm for computing the GMM barycenter under the MAW distance.

Main Results:

• The proposed MAW-based algorithm provides a computationally scalable solution for GMM barycenter calculation.
• Theoretical advances validate MAW as a suitable approximation for Wasserstein distance between GMMs.
• The MAW barycenter of GMMs shares the same expectation as the Wasserstein barycenter.
• The method demonstrates superior clustering performance on single-cell RNA-seq datasets compared to existing approaches.

Conclusions:

• The novel MAW-based approach offers an efficient and effective solution for integrating single-cell data clustering.
• This method advances the field of single-cell data analysis by improving cross-dataset integration.
• The algorithm's scalability and accuracy make it a valuable tool for researchers working with large-scale single-cell genomics data.