retrofit

retrofit decomposes cell type mixtures in spatial transcriptomics (ST) data using a Bayesian non-negative matrix factorization framework without requiring external single-cell expression references.

Key Features:

• Reference-free deconvolution: Decomposes cell type mixtures in ST data without using external single-cell expression references.
• Cell type proportion estimation: Estimates cell type proportions with improved accuracy compared to reference-based methods across varying spot sizes and sample heterogeneity.
• Gene expression reconstruction: Reconstructs gene expressions from spatial transcriptomics data.
• Robustness to reference quality: Performs irrespective of the quality or availability of single-cell reference data.

Scientific Applications:

• Spatial transcriptomics analysis: Decomposes cell type mixtures in ST datasets to identify spatial localization patterns.
• Mouse brain mapping: Recapitulates known cell-type localization patterns in Slide-seq datasets of mouse cerebellum without single-cell data.
• Cell type heterogeneity studies: Enables analysis of sample heterogeneity in spatial transcriptomics experiments.

Methodology:

Bayesian non-negative matrix factorization.