pharmaceutical industry may be spending $50bn a year on AI to speed up drug development. Most of the buzz revolves around AIs trained on biological data that could improve the hit-and-miss process of drug discovery. Drugs can take a decade to emerge, cost billions of dollars and succeed only 10% of the time.
Even a small improvement in speed and efficiency would be hugely valuable. But scientists have struggled to tame biological big data with conventional statistical tools. Machine learning makes it possible to sift through piles of information, from clinical patient data and genome sequences to images of body scans.
Last year DeepMind, an AI lab that is part of Google, made a breakthrough using its AlphaFold system to predict the structure of almost all proteins, which may one day help identify which molecules have therapeutic potential. Though only around a dozen drugs in development have so far involved the use of AI, the list may grow rapidly—especially for simple molecules with properties that are relatively easy to predict. In the case of these more straightforward chemistries, the future of medicine is looking ever more like a computational problem.
Jim Weatherall, who oversees data science and AI at AstraZeneca, says the technology is used in 70% of the British firm’s small molecules in development. Using a technique called “reinforcement learning", AstraZeneca’s AI is constantly tweaking its molecular suggestions and playing out how a tweaked molecule might react. Ali Mortazavi, boss of E-therapeutics, a biotech startup in London, says that knowing the sequences of all the genes in, say, the liver, lets his firm use software to design RNA molecules (which are more complex but, owing to their links to DNA,
. Read more on livemint.com