Researchers develop AI-based framework 'OncoMark' for predicting behaviour of cancer cell

Researchers have developed an AI-based framework 'OncoMark' that examines the internal processes a cancer cell employs to grow -- "reading the mind of cancer" -- in order to predict how it will behave.

The framework, described in a paper published in October in the journal Communications Biology, can help inform the understanding of tumour biology and support personalised cancer treatments.

The team from Ashoka University in Sonipat, Haryana, and S N Bose National Centre for Basic Sciences in Kolkata said currently, diagnosis of cancer rarely measures 'hallmarks of cancer'.

Outlined in a 2011 paper, the hallmarks of cancer refer to six internal biological processes which a normal cell employs to turn cancerous. The list was later expanded to include 10 processes totally.

Explaining the internal processes, Debayan Gupta, a lead researcher on the OncoMark paper, said, "In order so that we call a cell 'cancerous', what are the signs of it? For example, a cancer cell is going to look at its internal genetic signature and it's going to 'mess around with its death timer'."

"So, while a normal cell grows old and eventually dies, a cancer cell will turn off the death timer and say 'I'll just keep existing'," Gupta told PTI over a video call.

"Similarly, a normal cell has a growth suppressor. A cancer cell however turns off its growth suppressor and says 'I'm going to multiply as much as possible'," he said. Cancer cells are known to also employ similar biological processes in order to evade being attacked by the immune system.

However, while the hallmarks of cancer provide a foundation for understanding tumour biology, current diagnostic procedures rarely measure them.

"'Stages of cancer' are not reliable measures of how well a cancer patient is doing. What you're doing is you're looking at the cancer from the outside to know 'how big a tumour is', 'where it is', 'has it metastasised (spread to multiple locations)?'," Gupta said.

"So, you're really looking at things from the outside, measuring them, and then trying to come up with a view of how dangerous this cancer is," he added.

Using artificial intelligence (AI), OncoMark was trained on genetic data of 3.1 million cancer cells across 14 types to understand how hallmark processes, including immune evasion and metastasis, work together to promote tumour growth and resistance to treatment.

The AI-model was then tested on five independent datasets, consistently producing accuracy scores of minimum 96 per cent, the study says.

"We're looking at the potential of a cell to turn 'cancerous'. We're also looking at cells that are already cancerous and their potential to proliferate, potential to metastasise. And we're able to predict these very, very, very accurately, essentially with perfect accuracy," Gupta said.

"This can give doctors a lot more insight into sort of the internal, what we call 'reading the mind of cancer'," he added.

The authors wrote, "OncoMark calculates the probability of each hallmark's activity, providing a detailed molecular profile of the tumour."

The team is looking to get the model integrated into clinical workflows so that doctors can use it for patients.

The researchers also want to look into blood cancers, which have characteristics different to the standard solid cancers, and uncommon cancers.