One of the greatest risks to humanity is considered to be the evolution of bacteria resistant to available antibiotics. It is almost 100 years since Alexander Fleming discovered penicillin in 1928 and we now take for granted being able to easily treat diseases that just several decades ago could often prove fatal. But the recent rise of ‘super bugs’, that have evolved to be largely impervious to antibiotics has raised doomsday fears that a return to rates of mortality that existed before the current era of modern medicine couldn’t be discounted.

In that context, and even those of less dramatic but also unpleasant scenarios, recent news artificial intelligence has been used to discover new antibiotics that have proven effective against recently untreatable diseases, has been greeted enthusiastically.

There has been hope for some time that machine learning might prove a decisive new weapon in the arms race between antibiotics and drug-resistant bacteria. But last week’s publishing of a paper in the journal Cell, describing the discovery of halicin, a new antibiotic able to wipe out 35 powerful bacteria is a major landmark in the harnessing of artificial intelligence to develop new drugs.

Researchers at the Massachusetts Institute of Technology (MIT), used machine learning algorithms to develop halicin. Clostridium difficile, tuberculosis and Acinetobacter baumanni, an effectively untreatable infection of wounds that often causes death and has afflicted numerous U.S. military personnel, were all among the pathogens the scientists targeted with halicin.

James J Collins, the MIT biological engineer who led the research that has resulted in halicin, explained the significance of the machine learning-driven breakthrough:

“We are facing a global crisis, due to increased emergence of resistant bacterial pathogens that are rendering our current antibiotic arsenal ineffective. If we don’t address the crisis by 2050, the annual deaths due to antibiotic-resistant infections will grow to 10m, which is higher than the death rate due to cancer.”

Computer scientist Regina Barzilay developed the deep-learning algorithm that was trained to analyse the structure of 2500 molecules present in a combination of current antibiotics and natural compounds for anti-bacterial efficacy. 100 million molecules were then assessed by the AI, which predicted how potent they would each be when set against specific pathogens. It was then asked to filter for molecules with a different physical structure to existing antibiotics. Pathogens have evolved their physical structure to prevent known antibiotics from being able to latch on to them, rendering them essentially useless.

Ms Barzilay commented:

“There is still a question of whether machine-learning tools are really doing something intelligent in healthcare, and how we can develop them to be workhorses in the pharmaceuticals industry. This shows how far you can adapt this tool.”

However, it is not only advancements in the science of artificial intelligence and biotechnology that will influence how quickly new super antibiotics can be made available to doctors. There are also economic models to be overcome. As explained by Kevin Outterson, a healthcare lawyer for Carb-X, an NGO in the field of business models for antibiotics development:

“We see many promising new ideas against drug resistance. [But] the key gaps today are economic, not just scientific.”

One issue that faces investment in new, improved antibiotics is the fact that they are only used for several days at a time. Other drugs that involve similar levels of investment to develop are taken by patients over several weeks, months or even a lifetime. Pharmaceuticals companies are more incentivised to channel investment towards those kinds of treatments.

However, there is a growing body of evidence to support the supposition that using machine learning will significantly reduce the cost of discovering new drugs. As MIT’s Mr Collins says:

 “We could dramatically reduce the cost required to get through clinical trials.”