Scientists at Princess Margaret Cancer Centre (PM), in collaboration with researchers at the University of Pennsylvania, have created a new tool called AnnoSpat (Annotator and Spatial Pattern Finder) to identify different cell types and determine their position in tissues.
The cellular composition and spatial organization of cells in organs and tissues are crucial for proper function. Recently, advanced tools have been developed to measure cell types in tissues and their localization on a large scale.
“These tools, such as Image Mass Cytometry (IMC) and Co-Detection by Indexing (CODEX), enhance our ability to study the complex relationships between different cell types within tissues,” says Dr. Gregory Schwartz, Scientist at PM and co-senior author of the study. “However, the massive amount of data they produce creates a need for computational tools to accurately identify and map cell types.”
To address this need, the research team, also led by first author Dr. Aanchal Mongia and co-senior author Dr. Robert Faryabi at the University of Pennsylvania, created AnnoSpat. This tool uses a type of machine learning called a neural network, which models the structure and function of biological neural networks in the brain. It also employs a specific algorithm—based on point processes from probability theory—to quantify the spatial relationships among multiple cell types.
AnnoSpat can be used to the study how the microenvironment of organs can impact disease. For example, it can examine the cells in the pancreatic microenvironment during the development of Type 1 Diabetes (T1D).
“To test AnnoSpat’s accuracy and efficiency, we used it to identify different cell types and their distributions in pancreatic tissues,” says Dr. Schwartz. “We also found favorable performance when we compared its results with other methods of cell annotation for pancreatic tissue samples from both Type 1 Diabetes (T1D) and non-diabetic donors.”
To further assess AnnoSpat’s capability in analyzing the pancreatic microenvironment during T1D progression, the team studied pancreatic tissue from individuals with autoantibodies against pancreatic islet proteins in their blood but no clinical diagnosis of T1D.
After a comprehensive analysis of 1,170,000 cells from 143 slides of 19 donors in the Human Pancreas Analysis Program (HPAP), the team demonstrated that AnnoSpat can rapidly and accurately predict the identity of cells in tissue profiled with IMC and CODEX assays and capture the aggregation of immune cells with pancreatic islets in T1D samples.
Comparative studies further demonstrated that AnnoSpat can accurately predict the lineages of large fractions of cells, whereas other existing cell annotation algorithms often fail. AnnoSpat’s accuracy was also exemplified by its ability to correctly identify endocrine cells—those that produce hormones—that were mislabeled by expert annotation.
Overall, AnnoSpat is a powerful tool with the potential to help scientists better understand how different cells behave in tissues, leading to improved knowledge about diseases and potential treatment strategies.
Dr. Gregory Schwartz, Scientist at PM and Faculty Affiliate at the Vector Institute
This work was supported by the National Institutes of Health, the Canadian Institutes of Health Research, the Canadian Cancer Society, Human Islet Research Network, and Human Pancreas Analysis Program, and The Princess Margaret Cancer Foundation.
Dr. Gregory Schwartz is an Assistant Professor in the Department of Medical Biophysics at the University of Toronto and a Tier 2 Canada Research Chair in Bioinformatics and Computational Biology.
AnnoSpat and its individual components are available through https://github.com/faryabiLab/AnnoSpat.
Mongia A, Zohora FT, Burget NG, Zhou Y, Saunders DC, Wang YJ, Brissova M, Powers AC, Kaestner KH, Vahedi G, Naji A, Schwartz GW, Faryabi RB. AnnoSpat annotates cell types and quantifies cellular arrangements from spatial proteomics. Nat Commun. 2024 May 3;15(1):3744. doi: 10.1038/s41467-024-47334-0.