The covariance environment defines cellular niches for spatial inference

Area of Science:

• Computational biology
• Genomics
• Spatial transcriptomics

Background:

• Analyzing high-resolution spatial profiling data presents challenges in representing cellular neighborhoods.
• Understanding cellular interactions within these niches requires robust multivariate analysis.

Purpose of the Study:

• Introduce covariance environment (COVET) for representing cellular neighborhoods.
• Develop environmental variational inference (ENVI) to integrate spatial and single-cell RNA sequencing data.
• Enhance gene expression imputation and spatial context for single-cell genomics.

Main Methods:

• Covariance environment (COVET) leverages gene-gene covariate structure across cells.
• Optimal transport-based distance metric for comparing COVET niches.
• Environmental variational inference (ENVI), a conditional variational autoencoder, embeds spatial and single-cell RNA sequencing data.
• ENVI utilizes two decoders for gene expression imputation and spatial information projection.

Main Results:

• COVET captures multivariate cellular interactions within niches.
• The developed distance metric scales to millions of cells.
• ENVI successfully confers spatial context to genomics data from dissociated cells.
• ENVI outperforms existing methods for gene expression imputation on spatial datasets.

Conclusions:

• COVET provides a novel representation for cellular neighborhoods in spatial profiling.
• ENVI effectively integrates spatial and single-cell RNA sequencing data, improving analysis.
• The developed methods offer significant advancements for spatial genomics research.