Abstract:Spatial omics and multiplexed imaging technologies provide unprecedented spatial context to characterize molecular variation in complex tissues. Developing unsupervised computational pipeline in a discovery mode is of vital importance to interpret spatially resolved data and derive novel biological insights. We develop Stereo, a unified framework leveraging contrastive learning to jointly model molecular and spatial information in a self-supervised manner. Stereo consists of two modules, StereoCell for neighbor-aware cell identity inference and clustering on spatial transcriptomics data, and StereoNiche for biologically-relevant niche identification on spatial proteomics data. StereoCell uses graph attention to aggregate immediate neighbors’ content for each cell while StereoNiche exploits vision Transformer to explicitly relate position coordinates with marker readouts within certain receptive field. We validate StereoCell’s superior utility on both simulated and real transcriptomics datasets of brain tissues acquired by various technologies with different resolutions. StereoCell can operate on all genes, extract layer-specific signature genes at single-cell resolution and identify clear layer-wise structure with state-of-the-art performance. Applied on tumor tissue sections, StereoCell enables detailed characterization of intratumoral heterogeneity and reveals invasive or metastasis behaviors concordant with expert annotations. Evaluated on a multiplexed fluorescence imaging dataset for colorectal cancer (CRC) patients, StereoNiche refines niche assignments, recapitulates classical follicle structure, and validates enriched T cells-macrophages interdigitation. Via self-distillation operation, StereoNiche also enables extracting bounded layout of tertiary lymphoid structure with no supervision, representing classical Crohn’s-like reaction behavior. Local explanations are derived for a tree-based survival model to validate the clinical relevance of identified niches. For mass spectrometry profiled triple-negative breast cancer (TNBC) samples, StereoNiche recognizes differential tumor-immune interactions to separate compartmentalized from mixed pattern. In brief, Stereo is applicable to a variety of spatial technologies, serving as a powerful method to interrogate context-aware biological process and facilitate hierarchical signature discovery in neuroscience and computational pathology.

CellContrast: Reconstructing spatial relationships in single-cell RNA sequencing data via deep contrastive learning

conST: an interpretable multi-modal contrastive learning framework for spatial transcriptomics

A contrastive learning approach to integrate spatial transcriptomics and histological images

Graph Contrastive Learning of Subcellular-resolution Spatial Transcriptomics Improves Cell Type Annotation and Reveals Critical Molecular Pathways

STdGCN: spatial transcriptomic cell-type deconvolution using graph convolutional networks

Contrastive learning enables rapid mapping to multimodal single-cell atlas of multimillion scale

Position-informed contrastive learning for spatially resolved omics deciphers hierarchical tissue structure at both cellular and niche levels

A comprehensive comparison on cell-type composition inference for spatial transcriptomics data

Contrastive Learning for Robust Cell Annotation and Representation from Single-Cell Transcriptomics

Graph Contrastive Learning as a Versatile Foundation for Advanced scRNA-seq Data Analysis

ScCCL: Single-Cell Data Clustering Based on Self-Supervised Contrastive Learning

Identifying spatial domain by adapting transcriptomics with histology through contrastive learning

Predicting cell types with supervised contrastive learning on cells and their types

Deep enhanced constraint clustering based on contrastive learning for scRNA-seq data

spatiAlign: an unsupervised contrastive learning model for data integration of spatially resolved transcriptomics

MuCST: restoring and integrating heterogeneous morphology images and spatial transcriptomics data with contrastive learning

Multimodal contrastive learning for spatial gene expression prediction using histology images

A multi-view graph contrastive learning framework for deciphering spatially resolved transcriptomics data

LETSmix: a spatially informed and learning-based domain adaptation method for cell-type deconvolution in spatial transcriptomics

stDiff: a diffusion model for imputing spatial transcriptomics through single-cell transcriptomics