One difficult aspect of cell typing is discerning cancer vs. healthy cell populations. The "scMalignantFinder" package (https://github.com/Jonyyqn/scMalignantFinder) promises to automatically perform this classification in scRNA-seq data. They ID gene programs related to cancer, then implement a classifier (logistic regression, I think). However, it doesn't translate directly to spatial, though since panel sizes are smaller and platform effects are large. The spatial world needs a similar capability.
Clues to whether a cell / cluster is cancer:
- Expression of cancer programs called out in the scMalignantFinder paper, and of biologically relevant gene programs (proliferation, translation, glycolysis...)
- Higher transcripts per cell - cancers tend to be very transcriptionally active
- Positive PanCK immunofluorescence stain (standard in CosMx, not sure about other platforms)
- Perhaps other aspects of cell morphology
One difficult aspect of cell typing is discerning cancer vs. healthy cell populations. The "scMalignantFinder" package (https://github.com/Jonyyqn/scMalignantFinder) promises to automatically perform this classification in scRNA-seq data. They ID gene programs related to cancer, then implement a classifier (logistic regression, I think). However, it doesn't translate directly to spatial, though since panel sizes are smaller and platform effects are large. The spatial world needs a similar capability.
Clues to whether a cell / cluster is cancer: