Benchmarking sketching methods on spatial transcriptomics data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• High-throughput spatial transcriptomics (ST) generates massive datasets, leading to computational challenges.
• Existing sub-sampling methods (sketching) optimize for expression data but neglect spatial information, potentially introducing bias in ST analysis.
• This bias can distort tissue architecture by over- or under-sampling specific regions.

Purpose of the Study:

• To systematically evaluate the impact of different sketching strategies on spatial transcriptomics data analysis.
• To identify sketching methods that preserve both transcriptomic heterogeneity and spatial tissue architecture.
• To develop improved sketching approaches for large-scale ST datasets.

Main Methods:

• Benchmarking of uniform sampling, leverage-score sampling, Geosketch, and scSampler on diverse ST datasets (mouse ovary, MERFISH brain, human breast cancer, lung) and simulations.
• Evaluation using input representations including PCA embeddings, spatial coordinates, and spatially smoothed embeddings.
• Development and testing of a spatially aware sketching method using smoothed leverage scores from randomized SVD.

Main Results:

• Expression-only sketching captures global heterogeneity but distorts tissue architecture.
• Coordinate-only sketching maintains tissue coverage but misses transcriptional extremes.
• Spatially smoothed leverage scores effectively balance tissue coverage and transcriptomic representation, outperforming other methods in Hausdorff distance, clustering stability, PCA loading drift, and MSE.
• The proposed method recovers rare cell states and avoids edge effects.

Conclusions:

• Standard sketching methods are insufficient for unbiased spatial transcriptomics analysis due to their neglect of spatial information.
• A novel, spatially aware sketching approach using smoothed leverage scores offers a robust solution for analyzing large ST datasets.
• This method enables fast, accurate, and unbiased exploration of complex spatial transcriptomic data, preserving both tissue structure and cellular heterogeneity.