AI is accelerating both research and innovation. This is particularly noticeable in the pharmaceutical industry, where new medicines can now be developed from idea to preliminary clinical trials in less than 12 months, compared to the 4-5 years this took in the past.
Developing new medicines – and getting them approved – has not only typically been a time-consuming process, but also extremely costly. According to a report from Deloitte, the development cost of the medicines approved in 2018 was around SEK 20 billion on average.
– “Over the years, the cost has been driven up by the fact that it was not possible to determine early on which candidates are worth moving forward with in the process,” says Ian A. Cotgreave, Professor in Toxicology at RISE who has spent years in the pharmaceutical industry at, among others, Astra Zeneca. “To attain an approved medicine, you had to develop and test 10,000 candidates. With AI, the accuracy may be as high as one in ten instead. It is almost unbelievable how much AI can streamline the development process.”
‘Booms’ eliminated from the start
AI learns from millions of previous structures, experiments and tests, both successful and unsuccessful. Conclusions are also drawn from the entirety of the world’s available scientific literature in the field of medicine, from documents to patents, from medical records, from gene sequence data and so on. This means researchers do not need to waste time on courses that are doomed to fail.
AI warns of the candidates that could spell trouble.
– “It’s much better to abandon a course early than discover later on that candidates you believed in – and on which you invested 4-5 years of research – turn out to be completely ineffective on humans, or which may affect the heart or come with of life-threatening side effects,” says Cotgreave.
Computers can find candidates with the greatest potential.
– “It generates a heat map from which you can pick structures with the best chance of reaching the clinical phase,” he continues. “Previously, this first selection process took 12-16 weeks. Today it may take as little as a week.”
AI is unrivalled at analysing available data and establishing priorities, thereby speeding up processes. This is one of the explanations for the rapid development of the Covid-19 vaccine. However, it will be many years until AI is able to replace clinical validation. Because the human body is so complex – with different interconnected systems, which are all affected by both internal and external factors – not everything can be predicted.
Training the AI models takes time, of course, but afterwards it’s lightning fast
New materials can be developed much faster
Looking beyond the pharmaceutical industry, it is easy to see that, with the help of AI, research and innovation are progressing faster in virtually all areas of society. Another example is the development of new materials.
– “Perhaps most exciting is the possibility of AI replacing physics-based material simulation software,” says Andreas Thore, Researcher at RISE’s Department of Industrial Systems.
Calculating the three-dimensional atomic structure, or physical characteristics, of a single material with conventional software often takes several days. With AI, and access to large volumes of data from previous calculations and experiments, a simulation model can instead be trained to perform these calculations a million times faster.
– “Training the AI models takes time, of course, but afterwards it’s lightning fast,” says Thore. “However, it’s worth remembering that conventional, physics-based simulation software programs also took a very long time to develop.”
AI software can be used to design new materials for all kinds of application areas: batteries, solar panels, protective coatings, electronic components, and so on.
– “This means that the energy density of batteries and the efficiency of solar panels can be improved more rapidly. In turn, the cost per stored or generated unit of energy is falling at an ever faster rate and the transition to a fossil fuel-free society can be accelerated,” concludes Thore.
