Professor Nikolai Petrovsky, a pioneer in the field of immuno-informatics from Flinders’ University, claims to have developed the world’s first AI-generated drug to have entered human trials; and is encouraging other researchers to consider AI technologies for drug discovery also.
Deploying a range of AI technologies including Artificial Neural Networks (ANN), Hidden Markov models, Random Forrest Walk and Genetic Algorithms, Professor Petrovsky’s team tested whether AI could find a better drug for stimulating the innate immune system – specifically, toll like receptor 9 (TLR9) – than human experts have been able to design so far.
TLR9 serves the immune system by recognising DNA from pathogens (viruses and bacteria) and triggering an immune response against that pathogen. This ability makes drugs which activate TLR9 useful as vaccine adjuvants, that is, compounds that enhance vaccine effectiveness.
“We trained the AI by exposing it to examples of DNA sequences which had previously been shown to activate the receptor, together with examples of DNA sequences which don’t work. Through this repeated training the ANN ‘learned’ its own ways to identify DNA patterns that would potently stimulate TLR9”, said Petrovsky.
Using this approach, the AI identified completely new ligands of TLR9 for use as immune stimulating drugs and vaccine adjuvants.
“It took about six months for the AI to propose some suggested TLR9 active drugs it had designed. We got them synthesised and they proved to be extremely effective at activating TLR9.
“We then confirmed their safety and effectiveness as vaccine adjuvants in animal models. There wasn’t any toxicity; and from there we were able to move them rapidly into human clinical trials as both vaccine adjuvants and as anti-cancer compounds”, he said.
When asked how the AI was able to design TLR9 active drugs better than human experts, Petrovsky said, “This is the beauty of AI. While we know the theory of how it works in principle, we can’t tell afterwards exactly how it came up with the proposed solution.
“That’s the point of it. The technology has a ‘mind’ and ‘intelligence’ of its own. We give it the tools and the training and it then goes and figures the problem out for itself. It’s like a parent teaching a child how to read and that child goes onto to write a brilliant novel, something the parent themself had never conceived of when teaching the child.
“AI technology matures in the same way as a human. It acquires and accumulates knowledge and intelligence. It does the thinking, and looks for the complex patterns in the data so that humans don’t have to, or may not even be capable of”, he added.
The entire process took the researchers just four years, to go from drug discovery to human trials, a far cry from the average ten to twenty years it normally takes to bring a drug from initial lab screening to human trials.
“Normally a laboratory scientist would screen millions of compounds to get a useful lead which can take decades. The AI does this far more efficiently and, without stating the obvious, without requiring a salary in the process. The time and cost savings are incredibly significant”, Petrvosky added.
Drug discovery is a time- and capital-intensive process, and much of this investment goes to waste, with only one out of every ten therapies that reach human Phase 1 going on to gain regulatory approval.
Petrovsky says AI technology can potentially help shave decades – and hundreds of millions of dollars – from drug development; improving conversion rates as well as returns on investment for the drug industry.
Though the healthcare sector is not a stranger to AI – with the last few decades seeing a proliferation of AI technologies deployed in applications such as interpretation of ECG traces, treatment design, digital consultation and medical management – AI’s use in drug discovery and basic science research has been much more hesitant.
“There’s been a significant amount of resistance and skepticism amongst basic science researchers”, said Petrovsky. “I’m not quite sure why this is, particularly when you consider the low conversion rates and returns on drug R&D investment that currently exist. Few would doubt that we need a better approach and I believe that AI can provide exactly that”.
Twenty five years ago when doing his PhD, Professor Petrovsky and his colleagues were one of the first teams in the world to apply AI tools in immunology research, with Professor Petrovsky personally coining the term “immuno-informatics”, a term now widely used to describe the field.
Presenting at the AI, Machine Learning and Robotics in Health conference – due to take place 24-25 October 2019 – Professor Petrovsky will share further details of his latest AI-related drug discovery research.