Learning Predictive Models of Tissue Cellular Neighborhoods from Cell Phenotypes with Graph Pooling

Yuxuan Hu,Jiazhen Rong,Runzhi Xie,Yafei Xu,Jacqueline Peng,Lin Gao,Kai Tan

DOI: https://doi.org/10.1101/2022.11.06.515344

2022-01-01

Abstract:It remains poorly understood how different cell types organize and coordinate with each other to support tissue functions. We describe CytoCommunity for identification of tissue cellular neighborhoods (TCNs) based on cell phenotypes and their spatial distributions. CytoCommunity learns a mapping directly from cell phenotype space to TCN space by a graph neural network model without using additional gene or protein expression features and is thus applicable to tissue imaging data with a small number of measured features. By leveraging graph pooling, CytoCommunity enables de novo identification of condition-specific TCNs under the supervision of image labels. Using various types of single-cell-resolution spatial proteomics and transcriptomics images, we demonstrate that CytoCommunity can identify TCNs of variable sizes with substantial improvement over existing methods. To further evaluate the ability of CytoCommunity for discovering condition-specific TCNs by supervised learning, we apply it to colorectal and breast cancer tissue images with clinical outcome information. Our analysis reveals novel granulocyte- and cancer associated fibroblast-enriched TCNs specific to high-risk tumors as well as altered tumor-immune and tumor-stromal interactions within and between TCNs compared to low-risk tumors. CytoCommunity represents the first computational tool for end-to-end unsupervised and supervised analyses of single-cell spatial maps and enables direct discovery of conditional-specific cell-cell communication patterns across variable spatial scales.### Competing Interest StatementThe authors have declared no competing interest.