Randomized Spatial PCA (RASP): a computationally efficient method for dimensionality reduction of high-resolution...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) reveals gene expression patterns within tissue architecture.
• Understanding spatial domains is crucial for development and disease research.
• Existing ST analysis methods can be computationally intensive.

Purpose of the Study:

• Introduce Randomized Spatial PCA (RASP), a novel, fast, and scalable dimensionality reduction method for ST data.
• Enable flexible integration of non-transcriptomic data and de-noising of gene expression.
• Improve the efficiency of spatial domain identification and analysis in ST.

Main Methods:

• RASP employs a randomized two-stage principal component analysis (PCA) framework.
• Utilizes sparse matrix operations and configurable spatial smoothing for efficiency.
• Compares RASP against five existing methods on diverse ST datasets (10x Visium, Stereo-Seq, MERFISH, 10x Xenium).

Main Results:

• RASP demonstrates computational speeds orders-of-magnitude faster than existing techniques.
• Achieves comparable or superior performance in tissue domain detection.
• Effectively reconstructs de-noised and spatially smoothed gene expression values.

Conclusions:

• RASP offers a significant computational advantage for analyzing large-scale ST datasets.
• Facilitates exploration of high-resolution subcellular ST data.
• Enhances the study of tissue organization and biological processes through efficient spatial gene expression analysis.