One of every 31 Americans who enters a hospital will pick up an infection while there. That comes to about 1.7 million infections each year, resulting in 99,000 deaths, according to the U.S. Centers for Disease Control and Prevention.

What one researcher has called “public enemy No. 1 for multidrug-resistant bacterial infections” is Acinetobacter baumannii, a nasty bug that can cause meningitis, pneumonia, and other equally serious illnesses and was a chief cause of infections in wounded military personnel in Afghanistan and Iraq.

It lurks on hospital door handles and equipment and has proven stubbornly resistant to every pharmaceutical compound intended to kill it.

At the Massachusetts Institute of Technology, researchers used artificial intelligence (AI) and machine learning to sift through almost 7,000 possible formulas that might keep A. baumannii in check.

First, the research group grew the bacterium in lab dishes and exposed the bugs to about 7,500 compounds to see which ones had a negative effect on its growth.

Next, they told the AI the structure of each molecule that showed promise. Using machine learning, the AI then could recognize the features of a chemical that inhibits A. baumannii’s growth. 

Stuffed with that information, the AI then reviewed the chemical makeup of about 6,800 compounds it hadn’t seen before. 

Within two hours, the AI pulled out a few hundred good prospects. The research team chose 240 to test in lab dishes. Their choices were different from any known antibiotic and also from compounds the AI used to learn its task.

These “wet” tests yielded one candidate—an existing diabetes drug—that was especially lethal to the bacterium but had no effect on other bacteria, including the beneficial ones in the human digestive tract.

The drug’s focused killing ability also minimizes the risk of causing resistance to it among other species.

In lab tests, the drug killed A. baumannii in mice and also was lethal to strains of the bacterium taken from human hospital patients.

While they tweak their new drug to heighten its effectiveness, they also are making plans to use the same approach to find antidotes to other antibiotic-resistant bacteria.

TRENDPOST: Future drug discoveries could be even easier now that chemists at The Ohio State University have created G2Retro, an AI able to quickly predict how any group of two or more molecules will react together.

The AI was trained on 40,000 reactions of various molecules. 

Afterward, G2Retro was able to spin out hundreds of possible reactions for any given molecule within a few minutes.

With some coaching, G2Retro also was able to single out the molecules that would react best together to create a particular drug compound.

To test their AI in the real world, they decided to see if G2Retro could accurately predict the formulation of four drugs already in use for specific illnesses. 

It not only came up with the exact compounds, but also gave identical instructions for making them using methods now in use.

AI will allow the pharmaceutical industry to make stronger, more targeted drugs for a range of diseases and also will make it more financially feasible for the companies to produce relatively affordable drugs to treat rare conditions with small markets.