Machine learning identifies spatial signatures of kidney resident immune cells

ABSTRACT: Immune cells are spatially distributed in many organs and may be affected by the local microenvironment. The identification of spatial fingerprints that allow predicting the positioning of immune cells analysed by single cell sequencing (scRNAseq) would facilitate better understanding of their roles in health and disease. Here we aimed to identify such fingerprints by employing machine learning methods. We used the kidney as model organ because it can be divided into regions with distinct functions and microenvironmental cues, the cortex and the outer and inner medulla. We generated 3 scRNAseq datasets of immune cells by manually dissecting these three areas from healthy mouse kidney. Several machine learning algorithms including Neuronet, RandomForest, DecisionForest, multilayer perceptron (MLP) and others were utilized to identify genes harboring spatial information. Two external spatial datasets were used to validate the broad utility of the identified spatial marker genes. We found that the MLP algorithm identified a set of high variable genes that predicted the position of kidney-resident macrophages with accuracy of >75%. External validation substantiated the predictive power of this fingerprint. Gene-set enrichment analysis informed about the biological context of spatial marker genes. These marker genes were enriched in pathways relating to microenvironmental responses and cellular adaptation and showed a gender bias. Prediction was poor for motile immune cells like monocytes derived macrophages, T cells and B cells. Our algorithm even predicted the position of immune cells in an external spatial human dataset with comparable efficiency. Applying our strategy to an external spatial scRNAseq dataset of brain microglia allowed predicting the position of brain microglia. In conclusion, we identified a set of spatial marker genes that can predict the location of resident immune cells in the human and murine kidney and demonstrate that our strategy can be transposed to other organs.

ORGANISM(S): Mus musculus

PROVIDER: GSE262968 | GEO | 2026/01/01

REPOSITORIES: GEO