Despite some uncertainties around artificial intelligence in health care, one area where it is drawing significant interest is in drug discovery and development.
A partnership between Swiss drugmaker Novartis and US chip maker Intel has enabled researchers to use deep neural networks to cut the time for analyzing microscopic images from 11 hours to 31 minutes, Intel said last month. The announcement noted that the images used in the analysis are more than 26 times larger than those in a more commonly used dataset containing images of animals, objects and scenes.
Novartis isn’t alone in pursuing AI in drug discovery. British startup Exscientia, for example, has a collaboration with GlaxoSmithKline announced last year, shortly after it announced another partnership worth 250 million euros with Sanofi. The deal with GlaxoSmithKline, worth up to 33 million pounds in milestone payments, will see Exscientia’s platform used to discover small-molecule drugs for up to 10 disease-related targets selected by GSK. Pfizer, meanwhile, announced a deal in 2016 to use IBM Watson to speed up research on immunotherapy for oncology.
Indeed, the research Intel is doing with Novartis can be leveraged broadly across the industry for all types of drugs and compounds that involve very large images, said worldwide general manager for Intel’s health and life sciences group Jennifer Esposito, in an interview. The project used eight CPU-based servers and the dataflow programming library TensorFlow and was able to scale to more than 120 3.9-megapixel images per second. The fully trained model was able to process images with 99 percent accuracy, which is important given that the goal is to reduce human error and time in drug discovery, Esposito said.
The impetus for bringing AI and machine learning into drug discovery and development is obvious, as it often costs $2 billion and takes 12 years to bring a new drug to market. And that’s not counting the Mt. Everest-sized pile of drugs that never see the light of day because they fail preclinical and clinical testing.
AI and ML come with some real advantages over traditional methods. Combined with cryo-electron microscopy, or cryo-EM, researchers can use them to correctly infer three-dimensional structures of molecules photographed using cryo-EM and map them at the atomic level in ways not possible using traditional X-ray crystallography, BioPharm Insight reported in September 2017. Such novel methods will enable researchers to know much more about where on a target a drug must bind in order to be effective. BioPharm Insight reported that it will be two to five years before AI and ML become mainstream in drug development. Other potential applications include using AI to improve clinical trial design to decide whether to continue developing a drug and creating “synthetic comparator arms” to aggregate data from past clinical trials, the online news outlet reported.
A report last year by Global Market Insights found that the healthcare AI market – which includes drug discovery – was worth $750 million in 2016 and is likely to grow annually by 40% through 2024. While the report looked at healthcare AI broadly, it noted that drug discovery had more than 35 percent of the global market share and also was likely to grow by 40 percent during the same period.
Photo: Andrzej Woycicki, Getty Images