As scientists around the world race to see if they can come up with a vaccine for the coronavirus, one thing is clear to the experts at ML6: artificial intelligence can help.
ML6, a machine learning and artificial intelligence (AI) expert with offices in the Netherlands, Belgium, Germany and the UK, says its tools can help life sciences companies find new drugs to test, match patients with trials and improve quality controls in manufacturing highly-regulated medicines.
‘The playing field in which life sciences companies operate is getting ever harder,’ explains Simon Logghe, life sciences industry lead at ML6. ‘Typically, to develop a drug, test it and bring it to market can easily take more than 15 years for a pharmaceutical company. With patents only lasting 20 years, there’s a small margin to win the investment back. A lot of drugs don’t even get into the commercial phase as they don’t get through the clinical trial period. The consequence is that a lot of potentially life saving drugs are either very slow to get to market or don’t find the right patients for testing or have the right protocol for clinical trials.’
Currently, scientific teams around the world are searching for candidates that could treat the coronavirus, for instance. ‘One of the things they are trying to do is to find a cure for the coronavirus by identifying molecules that can expedite the formation of a treatment,’ says senior machine learning engineer Jeffrey Hagen.
‘One of the ways to do it is to use AI or machine learning. We can use generative adversarial networks that can create molecules and try to find the right one for a cure. These are similar networks to the ones you see in the news about deep fakes – a generative algorithm to manipulate visual and audio content with a high potential to deceive.’
‘If you did that same exercise in the real world, it would take millions of years to create a new drug. But new techniques that we use with our customers have shown to be very effective in these complex environments and can increase the success rate of the drugs that they find.’
Another way in which pharmaceuticals, bio-tech and med-tech companies can use machine learning is in matching potential drugs with candidates for clinical trials around the world. This is a technique which ML6 has already used in helping an HR company to find new candidates for roles, based on matching people’s skills with requirements.
‘If you develop a specific cancer drug, where are the patients I need to run my clinical trial?’ says Logghe. ‘The data is sitting in all hospitals, with all physicians but they write their notes in the computer in full text. We can use AI to read the physician’s text and make it structured according to the disease, diagnosis, and socioeconomic information about the patient.
‘Then you can use it to find where the patients are. For some drugs it can take multiple years to find your first set of 10 to 20 patients. With AI and machine learning, you can do this in seconds.’
Privacy concerns can also be sidestepped by using anonymised data in a ‘latent space’ – so that actual patients are only ever looked at and contacted if they are a potential match. ‘The idea is to input text and then represent words as vectors in a space with multiple dimensions,’ says Logghe. ‘Word vectors which share common contexts will be close to another.’
This technique can be applied to all kinds of things. ‘For example, if you are looking for a candidate in an HR company, you can look for how many years’ experience someone has, and other qualities from a CV,’ adds Hagen. ‘You create a point in the space and cluster certain kinds of profiles together. You do the same for the job opening. If you plot them both, you can find the right candidate for the right job by looking at this latent space. You can do the same thing for patients and drugs.’
A string of numbers – rather like a DNA string of genetic information – compares the ‘ideal patient’ for a drug trial with actual patients around the world, also looking for hidden patterns that a doctor or human couldn’t see or correlations in huge data sets.
Computers can also help to define the protocols for a drug trial, and how it will work, by scanning and looking at similar trials for other successful medicines. ‘The protocol describes the way you are going to execute your trial, the target patient populations – like females, non-smokers, the minimum and maximum ages,’ says Logghe.
‘These are all variables you define in your protocol that have a very big impact on the outcome. Research shows that just by creating a wrong protocol can result in a non-effective trial.
‘By looking at vast amounts of historic research papers you can really understand where similar experiments have been done, what measures they used and whether they were successful: full text data where we use Natural Language Processing (NLP) algorithms.’
Manufacturing is another area where machine learning is critical, especially with highly-regulated drugs like narcotics or psychotropics. ML6 has been working for one of the world’s largest pharmaceutical companies using the ability of computers to understand visual images.
‘Samples are taken from the production line and we look at hyperspectral images – colours including infrared and other light frequencies to judge the product in terms of quality,’ says Jeffrey.
‘In the past it had to be done manually but we have created an algorithm that does it automatically for every medicine or every pill you are producing. This allows for much more quality control, ensures patients are getting safe drugs and you have full traceability within your plant.’
Meanwhile, life sciences companies are also considering using computer power to understand more about their drugs in the patient journey: encouraging people to take medicines when they are supposed to – perhaps using smart devices – getting feedback and using this to improve their medicines. One thing that does need to happen, the ML6 experts say, is for data to be shared in a more centralised and standardised way rather than stored only in separate systems in separate hospitals.
‘AI can help a lot but progress needs to come on two fronts: collaboration and standardisation,’ says Logghe. ‘The response to the coronavirus is a great example of how the urgency of finding a cure goes beyond the traditional protectionism, with researchers who are opening up their insights, data sets and platforms. It is showing a shift in the paradigm and this is going to continue.’