A research team from McGill University has developed an artificial intelligence tool capable of detecting the small groups of cells that contribute most to the progression of aggressive cancers.
Named SIDISH, this tool helps scientists design targeted treatments by highlighting the tumor cells most closely associated with poor patient outcomes. It avoids treating all cancer cells as if they behave the same way.
In a preclinical study published in
Nature Communications, the SIDISH method identified "high-risk" cells in tumor samples taken from patients with pancreatic, breast, or lung cancer and analyzed in the laboratory.
How the tool works
The main novelty of the SIDISH method lies in its ability to link what happens inside cells to patient outcomes, a long-standing quest in cancer research.
"On one hand, single-cell data are very detailed, but they usually come from only a small number of patients and rarely provide an indication of actual health outcomes. On the other hand, patient data, often analyzed globally, contain survival-related information, but they represent an average of signals from millions of cells, preventing the identification of rare but dangerous cells that drive the disease," explains Yasmin Jolasun, a doctoral student in the Department of Medicine at McGill University and lead author of the study.
It is difficult to effectively combine these two types of data using current computational tools.
"Our tool bridges these two worlds. Indeed, it detects cells most closely linked to rapid disease progression and provides insights into patient survival," says Yasmin Jolasun.
She adds that although the SIDISH method was first tested in cancer, it could be used for other complex diseases where differences between cells play an important role.
SIDISH stands for "Semi-supervised Iterative Deep Learning for Identifying Single-cell High-Risk Populations."
Predicting therapeutic targets before lab trials
In addition to identifying the problem, the SIDISH tool can simulate how high-risk cells would respond to the activation or deactivation of certain genes, helping to determine which genes might be promising therapeutic targets.
"Since it usually takes years of trial-and-error research to find the right targets, this approach could accelerate drug development," says Jun Ding, senior author of the study, an assistant professor in the Department of Medicine at McGill University, and a junior scientist at the Research Institute of the McGill University Health Centre.
For example, a patient's tumor would be analyzed using single-cell sequencing, then the SIDISH tool would identify the cells driving the tumor, simulate their response to different drugs, and produce a shortlist of the treatments most likely to be effective, the researcher explains.
"In the short term, the SIDISH method could allow us to find new indications for existing FDA-approved drugs by leveraging public datasets. In the long term, the tool could radically change the drug discovery process," says Jun Ding.
The tool is still in development and is not yet used in clinical practice. The research team is attempting to apply the SIDISH method to other diseases and is collaborating with partner companies to refine it.