Dolphin AI reveals hundreds of invisible cancer marks

In a study published in Nature CommunicationsThe researchers explain how doctors can one day use doctors one day to capture diseases early and to direct treatment options.

“This tool has the ability to help doctors to match patients with treatments most likely to work for them, which reduces experiences and error in treatment,” said the author of the author John Ding, an assistant professor at McGill’s medicine and a young scientist at the Research Institute at McGill University Health Center.

Zoom in genetic building blocks

Signs of the disease are often hidden changes in the expression of RNA which can indicate the time of the disease, the extent of severity or how it can respond to treatment.

The researchers said that the traditional methods of analysis at the level of genes these signs collapse in one number per gene, which hides the critical contrast and capturing the tip of the ice only.

Now, making progress in artificial intelligence can capture the exact complexity of one cell data. Dolphin moves beyond the level of genes, as it enlarges to see how genes are divided together from smaller pieces called Exons to provide a clearer vision of cells.

“Genes are not just one block, they are like LEGO collections made of many smaller pieces,” said first author Kayo Song, a PhD student in the quantum life science program in McGil. “By looking at how to connect these pieces, our performance reveals the important signs of the disease that has been ignored long ago.”

In one of the test cases, the dolphin analyzed one cell data of pancreatic cancer patients and found more than 800 diseases missed by traditional tools. I have been able to distinguish between patients with high -risk and aggressive cancers from those with less severe cases, which would help doctors choose the path of the correct treatment.

A step towards “virtual cells”

On a larger scale, the penetration is the basis for achieving a long -term goal of building digital models for human cells. Dolphin generates richer profiles than traditional methods, allowing hypothetical simulations how cells and responding to drugs before moving to laboratory or clinical trials, saving time and money.

The next step for researchers will be to expand the tool from a few data collections to millions of cells, paving the way for the most accurate virtual cell models in the future.

About study

“Dolphin develops a single -cell transcriptomics outside genes by taking advantage of EXON and Junction readings” by Kailu Song and Jun Ding Et Al.Nature Communications.

This research was supported by the Mekins-Cristie Chair in respiratory research, Canadian Institutes of Health Research, Natural Sciences, the Canada Engineering Research Council, and Fonds de Rehashche du Québec.