Method eliminates the need for manual identification of cell types

Tumours are more than just a clump of cancer cells. Within them you will also find stroma cells connecting tissue, various immune cells, and other types. Understanding this cellular makeup is essential to many areas of cancer research, especially biomarker discovery. Recently, a team of Toronto-based researchers unveiled a new AI-based method of analyzing data from imaging mass cytometry (IMC) and other high parameter technologies used to determine a sample’s cellular profile. Their work is described in a recent article in the journal Cell Systems.

“One of the big questions in this field is how do you go from the raw data generated by highly multiplexed imaging such as IMC to actually determining the presence and abundance, or absence, of different cell types,” explains the study’s lead author Dr. Kieran Campbell, Investigator at Lunenfeld-Tanenbaum Research Institute and an OICR Affiliate researcher. “Right now, if you take a tissue section and run it through IMC there is quite a laborious, manual process of analyzing the resulting data that is open to subjectivity due to the need for human annotators to assign cell types.”

IMC provides a high-dimensional view of a tumour sample by mapping the expression of about 40 different proteins. Analysis of the data allows for single cells within a sample to be assigned to a type. To address the shortcomings of current methods of analysis Campbell and his collaborators created “Astir” (ASsignmenT of sIngle-cell pRoteomics) to automate the process of annotating cells, doing away with the need for human annotators and offering other advantages.

“Current methods of automation can only go as far as putting individual cells into groups and then it’s off to the human annotators to determine what the cells in those groups are. However, with Astir we can automate the assignment of cell types for each individual cell within a sample based on previously known biology,” explains Campbell. “Astir is also able to do this at great speed – in our study we showed that the system could assign a cell type to 800,000 cells in 15 minutes on a standard desktop computer.”

Campbell and his collaborators, which include Dr. Hartland Jackson, OICR Investigator, envision two primary uses for this new technology. In a discovery setting, researchers could use Astir to conduct an initial broad analysis of a dataset to get a high-level picture of the cellular makeup of their samples. In addition, with further development, the research group sees future clinical use for their invention to look for predictive or prognostic cell types within a single sample. “This tool is crucial for both our ability to analyze millions of cells from each clinical sample we investigate and within these samples identify rare cells that may play an important role in cancer development or progression,” explains Jackson.

“Infiltration of a tumour by immune cells is a prognostic indicator across many cancers, but to infer this from IMC currently requires a lot of manual analysis and expertise,” explains Campbell. “So this is really the first step towards automating that. We look forward to continuing development to include elements such as the spatial location of the cells within a sample with the goal of enabling new discoveries and ultimately helping patients.”