Although uncertainties exist around AI in healthcare, drug discovery and development is one area where the technology draws significant interest.
Analyzing microscopic images has been cut down from 11 hours to 31 minutes using deep neural networks, with the support of a partnership between Swiss drugmaker Novartis and US chipmaker Intel. The images used in the analysis, are 26 or more times larger than the images in a more commonly used dataset having images of objects, animals, and scenes.
This research by Intel, with Novartis, can indeed be utilized broadly across the vertical for every type of drug and compound that involves too large images. This project which leveraged eight CPU-based servers and the dataflow programming library TensorFlow scaled to more than 120 3.9 megapixel images per second. Given that the aim is to reduce human error and time in drug discovery, this fully trained model processed images with 99 percent accuracy.
One of the several real advantages of AI and ML over traditional advantages is that researchers can use them, combines with cryo-electron microscopy, or cryo-EM, to correctly infer three-dimensional structures of molecules photographed and achieve what is not possible with traditional X-ray crystallography— mapping them at the atomic level. Such means provides researches with more knowledge about the right place on a target, a drug must bind, for it to be effective. Several other potential applications include using AI to enhance clinical trial design to decide on the continuation of developing a drug and creation of “synthetic comparator arms” to gather data from past trials.
The healthcare AI market is likely to grow annually by 40 percent through 2024, from $750 million in 2016, said a report from the last year by Global Market Insights. Although the report broadly looked at AI in healthcare, it marked that drug discovery owned more than 35 percent of the global market.