Understanding disease using AI: The ultimate matchmaker

Ian Churcher


I spent a fascinating day at the Royal Society of Medicine (RSM) meeting on ‘Recent Developments in Digital Health’ held at their Marylebone headquarters.

It was a chance to hear about the increasing number of ways technology and artificial intelligence is helping to improve the care provided to patients. From better diagnosis by GPs in primary care; to more accurate image analysis to help radiologists identify breast tumours, and virtual reality techniques to help train surgeons.  The pace of change is fast.

It was inspiring to see the range of ways the medical community is now embracing technology to help patients, but it was also sobering to realise that many conditions and diseases still do not have effective treatments which will work for all, or even some patients.

Given this continued overriding medical need, I was appreciative of the opportunity to give an update on how BenevolentAI is using AI, machine learning and data from many diverse sources to answer some really difficult questions which underpin the search for new medicines: i) finding the best drug targets and therapeutic hypotheses; ii) identifying the right molecule to test in the clinic and iii) matching these medicines to the patients who are most likely to gain benefit from them.

This latter question, understanding disease to match the right treatment to the right patient, (often referred to as personalised or precision medicine) is a field not without controversy. After the sequencing of the human genome in 2003, much excitement ensued: being able to read an individual’s genes would unlock limitless knowledge on the causes of disease and surely allow us to discover new drugs tailored to patients’ specific circumstances?

Wind the clock forward 15+ years however and this hasn’t really happened outside of a few cases, mainly in oncology (where the genetic driver of disease may be more direct). Maybe this shouldn’t be a surprise to anyone. What went wrong? Nature vs nurture? Identical twins have very different susceptibility to disease. Genes are part of the story but only a part: using genetic information alone to understand disease is only part of the picture.

We’ve been trying to understand disease and personalised medicine in other ways for a long time of course: Listening to the discussion amongst some GPs at the conference, it’s obvious that personalised medicine analysis goes on in every patient consultation in every doctor’s office. A physician looks at the patient in front of them, assesses their symptoms, their family history, other risk factors they can identify, perhaps results from some tests and then makes a recommendation of an individual treatment plan based on this – a personalised medicine plan based on experienced medical observation and judgement. In helping understand disease, this is also a part of the story but, again, only a part.

Truly understanding a patient’s illness and finding new ways to treat it requires an understanding of all of these factors; their symptoms, their history, the underlying causes, their genome, its expression and epigenetic control, underlying network biology and many other factors. Understanding and reasoning across all of those domains is really difficult and in a world of deep academic specialisms it’s rare to find experts who are truly fluent in all these areas.

This is where we need to bring the power of AI to help researchers draw meaning from impossibly complex and disparate datasets to help in the search for medicines to treat disease. This is what we are setting out to do – bringing together the insight from genetic, biological and medical fields which often don’t lend themselves to easy integration.

Much of the information we need to help develop new medicines is out there, hiding in plain sight. AI can be the ultimate scientific matchmaker, bringing together the pieces of data that produce new insight but which just never had the chance to meet before.

With the vast amounts of new data becoming available, there’s never been a more exciting time to discover new drugs… if we have the help of the right matchmakers to bring the most compatible data together.



Blog

Using Artificial Intelligence to Optimise Small-Molecule Drug Design