CellART: a unified framework for extracting single-cell information from high-resolution spatial transcriptomics

CellART is a novel framework that unifies the process of extracting single-cell data from high-resolution spatial transcriptomics (ST) platforms. This research, published in the field of biomedical and biological sciences, addresses the challenge of converting sparse transcript counts into comprehensive single-cell information. Traditional SP platforms often yield limited gene measurements per spot, hindering a detailed analysis.

CellART overcomes this by amalgamating multimodal data, such as staining images and single-cell RNA sequencing references, using deep learning and probabilistic modeling. This integration enables both cell segmentation and type annotation in a single step. The framework's effectiveness is showcased across multiple high-resolution ST platforms, including VisiumHD, Xenium, MERFISH, and Stereo-seq, demonstrating its capability to process datasets containing millions of spots.

Rigorous experiments validate the framework's biological relevance and accuracy, particularly in the context of cancer and immune cells. Notably, CellART highlights its utility in uncovering transient cancer cell states, immune cell subtypes, and novel cancer-immune cell interactions. The outputs of CellART are compatible with widely used community tools, thereby facilitating a wide range of downstream analyses.