Harmonic Alignment.

Area of Science:

• Computational biology
• Data science
• Bioinformatics

Background:

• Integrating diverse biological datasets is challenging due to varying modalities and batch effects.
• Existing methods often require point-wise correspondence, limiting their applicability.
• Preserving intrinsic data structure during integration is crucial for accurate downstream analysis.

Purpose of the Study:

• To propose a novel framework for dataset integration via intrinsic geometric alignment.
• To enable fusion of data from disparate modalities and correct batch effects.
• To develop a method that does not require point-wise correspondence between datasets.

Main Methods:

• Alignment of intrinsic data geometry using diffusion operators and harmonic expansions.
• Leveraging partial feature correspondence to establish isometric alignment.
• Construction of a unified diffusion geometry over aligned datasets.

Main Results:

• Demonstrated effective fusion of single-cell RNA sequencing (scRNA-seq) and single-cell ATAC sequencing (scATAC-seq) data.
• Successfully removed batch effects between biological samples while preserving intrinsic data structure.
• The proposed framework shows robustness across various datasets.

Conclusions:

• The novel framework provides a powerful approach for integrating heterogeneous biological data.
• Geometric alignment based on partial feature correspondence offers a flexible alternative to point-wise matching.
• This method has significant implications for advancing multi-modal single-cell data analysis and batch effect correction.