University team uses AI to decode asbestos-linked cancer

Researchers at the University of Leicester have used artificial intelligence (AI) to gain new insights into mesothelioma, a cancer caused by breathing asbestos particles that typically occurs in the linings of the lungs or abdomen.

The Leicester Mesothelioma Research Programme used AI analysis of DNA-sequenced mesotheliomas to reveal that the tumours evolve along similar or repeated paths between individuals.

The research has now been published in Nature Communications. These paths predict the aggressiveness of the disease and thus possible therapies for this otherwise incurable cancer.

Leicester Mesothelioma Research Programme director Professor Dean Fennell said: “It has long been appreciated that asbestos causes mesothelioma, however how this occurs remains a mystery.

“Using AI to interrogate genomic ‘big data’, this initial work shows us that mesotheliomas follow ordered paths of mutations during development, and that these so-called trajectories predict not only how long a patient may survive, but also how to better treat the cancer—something Leicester aims to lead on internationally through clinical trial initiatives.”

Fennell recently made a major breakthrough regarding the treatment of mesothelioma in collaboration with the University of Sheffield, demonstrating that use of an immunotherapy drug called nivolumab increased survival and stabilised the disease for patients.

This was the first-ever trial to demonstrate improved survival in patients with relapsed mesothelioma, for whom chemotherapy has historically been the only treatment option, Currently, only 7% of people survive five years after a mesothelioma diagnosis, with prognosis averaging 12 to 18 months.

While the use of asbestos has been outlawed in the UK since 1999, cases of mesothelioma have increased by 61% since the early 1990s. AI disease prediction models like the Leicester team’s project may still have a long way to go before they become routine clinical practice.

Research from the University of Manchester recently found that the predicted risks for the same patients were very different between machine learning models and QRISK, a popular statistical tool for assessing patient risk of cardiovascular disease, particularly for high risk-patients.

Different machine learning models also gave different predictions, and were not able to take what statisticians call ‘censoring’ into account: patient data statistics move around, skewing the calculations downwards.

Last year, Australian researchers published findings that eligibility criteria for mesothelioma clinical trials may be too strict, with only a maximum of 63% of the team’s mesothelioma patients found to be eligible for two key trials studying treatments for the disease.