RCTD

Robust Cell Type Decomposition (RCTD) is a computational method for inferring cell type composition in spatial transcriptomics data by leveraging a reference single-cell RNA-seq dataset. The approach treats each spatial location (spot) as a mixture of cell types, modeling the observed expression as a weighted sum of cell type–specific signatures drawn from the reference.

RCTD uses a likelihood-based, robust estimation framework to estimate the mixture proportions for each spot, producing per-spot cell type proportions and, in some configurations, deconvolved expression profiles. The method incorporates gene selection and normalization to align reference profiles with spatial measurements and is designed to be resilient to differences between single-cell references and spatial data, reducing sensitivity to outlier genes.

Implementation and availability: RCTD has been released as an open-source software package, commonly used within the R programming ecosystem. It provides functions to input a reference scRNA-seq matrix with annotated cell types and a spatial transcriptomics count matrix, then compute estimated cell type proportions for each spot and optional deconvolved expression signatures.

Applications: The method has been applied to a range of tissue types, including brain and tumor microenvironments, to reveal spatial patterns of major cell types and subtypes within tissue sections. It is frequently used with platforms that generate spot-level expression data, such as Visium, to interpret tissue architecture.

Limitations and considerations: The accuracy of RCTD depends on the quality and completeness of the reference scRNA-seq dataset; rare or missing cell types may be misrepresented. The approach assumes that spot expression arises from a fixed set of reference cell types and may be sensitive to parameter choices and normalization procedures. Computational demands can be nontrivial for large datasets.

See also: deconvolution methods for spatial transcriptomics such as SPOTlight and other reference-based approaches.