Discovering new drugs to treat cancer has to be one of science’s highest callings. It’s certainly one of the most urgent — and potentially lucrative. More than 400,000 people in the UK receive a cancer diagnosis each year.
However, while an ageing population means more cases than ever before, mortality rates are not rising at the same pace, thanks to the slow but steady flow of novel drugs approved each year. Targeted therapies, which act on cancer cells or the specific proteins that help them grow while sparing the healthy cells, are becoming increasingly popular.
Frustratingly, it can take between 10 and 12 years on average for a single drug to move through the process of the research and development phase before approval. But new AI technology developed by a team at The Institute of Cancer Research (ICR) in London could change that. Called MorphoMIL, it uses deep learning to profile the shape of drug-treated cancer cells in 3D and check for changes.
This is a previously untapped reservoir of information
Previous research has focused on cancer cells in 2D, as they appear on microscope slides. Geometric deep learning, unlike a traditional convolutional neural network (CNN), transforms the input to a sphere that can be rotated. Applying machine learning in this way has allowed researchers to model these cells as they appear in the human body, and to better understand how the shape of the cell relates to its current state.
Professor Chris Bakal, professor of cancer morphodynamics at the ICR, likened the process to finding cell fingerprints. “3D cell shape is like a fingerprint of cellular state and function — it’s a previously untapped reservoir of information,” he explained. “Using AI, we can decode this fingerprint and reveal how cells respond to drugs.”
We will be able to streamline the years-long drug discovery process, saving both time and money
Researchers trained the AI tool on imaging of 95,000 melanoma cells treated with a variety of drugs so it could learn which cell shape changes had been caused by which drug. They reported that MorphoMIL could predict this with 99.3 per cent accuracy, and was able to identify proteins that could be targeted in developing new drugs.
The ICR believes it could slash preclinical research phases from three years to three months, help identify patients who would most benefit from a particular drug and predict side effects — potentially taking six years off the process of drug trials. “The tool we’ve created is so powerful that we will be able to streamline the years-long drug discovery process, saving both time and money,” said Professor Bakal. “Patients with cancer need new treatment options as quickly as possible, so speeding up this process will be hugely valuable.” Dr Bakal is now the chief scientific officer of Sentinal4D, a precision oncology company that has spun out of the ICR to focus on developing and applying this kind of patented AI technology.
It will be possible to predict how effective a drug will be and if there are likely to be any side effects
Sentinal4D co-founder Dr Matt de Vries said their technology would help cut out guesswork and improve success rates during drugs trials. “With the AI tool we’ve created, it will be possible to predict how effective a drug will be and if there are likely to be any side effects,” he said. As the study also examined red blood cells, he says it could be applied outside of the cancer research space. “The tool could work for a range of diseases, as we’ve shown that it will pick up the changes in shape for a number of different cell types and drugs.”
Sentinal4D has completed its first round of pre-seed funding, and excitement is high. “This technology builds on years of work at the ICR to understand cancer cell shape and to use artificial intelligence to analyse data,” said professor Keith Helin, the ICR’s chief executive. “I look forward to seeing this being used to develop new medicines that have a real impact for people with cancer.”
