The last year has seen some tremendous progress made in our ability to understand images, both still and moving, using artificial intelligence.  This has some obvious benefits in areas such as security and surveillance, but one startup believes it can also help us hunt for novel medicines.

The startup, called Recursion Pharmaceuticals, believe that by using a combination of robotics and machine vision, they can work simultaneously on hundreds of diseases.

The company have developed software that automates the testing of drugs inside the cells.  The software is clever because it uses machine vision to inspect the cells at a high level of detail.  It’s capable of measuring thousands of features of each cell, whether it’s the size or the structure.

Automated sleuths

The software is looking for cells with telltale signs that can make sick cells appear healthy.  They’re doing this across over 2,000 compounds at the moment.

The method has already identified 15 treatments that show promise for a number of rare diseases, which because of their relatively small population sizes often lack approved drug treatments.  With the cost of drug development so high, it usually isn’t economic to develop treatments for markets under 200,000 people in size.

One of these 15 is due to enter clinical trials later this year, and aims to treat cerebral cavernous malformation, which currently affects 60,000 people in the United States.  The plan is for a further three treatments to enter trials next year.

Recombination

The approach mirrors the recombinative approach to innovation, whereby innovators look for what already exists and explores new ways of applying it.  Recursion work with known drugs and try to find new uses for them.

The company has secured $19 million in funding and their core software originates from the Broad Institute, which is a collaborative lab between MIT and Harvard University.  The lab specializes in extracting information from biological images.

As with so many automated use cases in healthcare, the method allows potential treatments to be identified quickly and affordably, thus opening up exploration into treatments that are currently not economically viable.

Of course, the concept is at an early stage, and therefore still needs to prove that it can make the production of marketable treatments considerably easier to find.  The challenge remains to understand the inner workings of diseases, and how these can be altered.

The company plan to partner with a number of pharma companies to help with this process, and to develop the treatments that are identified.  They also hope that the data they generate will itself be a valuable resource over time.  It’s certainly an interesting company, and further evidence of the interesting strides being made when data and AI converge in healthcare.