CORAL: Multi-scale Multi-modal integration of Spatial Omics via Deep Generative Model

CORAL is a probabilistic, graph-based method designed to integrate diverse spatial omics datasets. Taking multimodality molecular profiles of unmatched spatial resolution and detected features, CORAL generates single-cell embedding with information from both data modalities, deconvolves the lower-resolution modality to infer its profile in individual cells, and predicts interactions between neighboring cells.

Installation

pip install git+https://github.com/zou-group/CORAL

Model Input (h5ad format):

• A hires adata
• A lowres adata
• (optional) the major cell type annotation on hires adata
• (optional) ground truth adata

Accepting spatial omics data

• spatial transcriptics
• spatial proteomics
• spatial metabolics
• spatial ATAC

Features

• Incorporating the scale-invariant feature transform (SIFT) to align the adjacent slides
• Generating joint single-cell embedding with two data modalities
• Deconvolves the lower-resolution modality to higy-resolution
• Inferring spatial niches
• Prediciting spatial variables
• Predicting interations between neighboring cells

simple core implementation

import coral
combined_expr, hires_coords, one_hot_cell_types, spot_indices, lowres_expr = coral.utils.preprocess_data(lowres_adata, hires_adata)
    
dataloader = coral.utils.prepare_local_subgraphs(combined_expr, hires_coords, one_hot_cell_types, 
                                           spot_indices, lowres_expr,n_neighbors=40)    

model, optimizer = coral.model.create_model(lowres_dim = lowres_adata.shape[1],
                                            hires_dim = hires_adata.shape[1],
                                            lowres_size = lowres_adata.shape[0],
                                            hires_size = hires_adata.shape[0],
                                            cell_type_dim=one_hot_cell_types.shape[1],                                          
                                            latent_dim=64, 
                                            hidden_channels=128, 
                                            v_dim = 1
                                          )
coral.trainer.train_model(model, optimizer, dataloader, epochs = 100 ,device = device)
adata_model_gener = coral.inference.generate_and_validate(model, dataloader,device, hires_adata)

Tutorial

• CORAL provides semi-automated alignment using SIFT. If your spatial multi-omics data are not aligned, please check our tutorial for alignment
• After the two input adata are prepared, please folowing our basic tutorial
• For additional steps related to reproducing the results in the paper, please refer to our reproducibility repository: https://github.com/siyuh/CORAL_reproducibility

How to cite CORAL

bioRxiv: https://doi.org/10.1101/2025.02.01.636038

Contact

In case you have questions, please contact:

• Siyu He - siyuhe@stanford.edu
• via Github Issues