CellPie: a scalable spatial transcriptomics factor discovery method via joint non-negative matrix factorization.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (ST) preserves gene expression data within tissue context.
• Histopathological images offer complementary morphological information to ST data.
• Integrating these data types can enhance biological insights.

Purpose of the Study:

• To introduce CellPie, a novel unsupervised method for joint factor discovery from spatial transcriptomics and histology data.
• To develop a computationally efficient approach for analyzing high-dimensional ST datasets.
• To demonstrate the performance and speed of CellPie across diverse cancer types and spatial resolutions.

Main Methods:

• CellPie utilizes joint non-negative matrix factorization (NMF) on spatial RNA transcripts and histological image features.
• The method incorporates an accelerated hierarchical least squares algorithm to reduce computational time.
• The approach was validated on three human cancer types, including Visium HD datasets.

Main Results:

• CellPie achieves high computational efficiency, significantly reducing analysis time for ST datasets.
• The method demonstrates good performance in factor discovery by integrating transcriptomic and imaging data.
• Validation across different cancer types and resolutions confirms CellPie's robustness.

Conclusions:

• CellPie offers a fast and effective unsupervised approach for integrating spatial transcriptomics and histology.
• The method's computational efficiency makes it suitable for large-scale, high-dimensional ST data analysis.
• CellPie enhances the study of gene expression within tissue architecture, advancing cancer research.