AI in Discovering Powerful Antibiotics

AI’s recent approach helped the researchers to discover new types of antibiotics that are powerful. The team did this as machine learning can analyze 100 million molecules precisely. From focusing on a variety of bacteria such as tuberculosis and strains to examining them, machine learning becomes successful in finding a proper antibiotic.

Language Problem in Antibiotic Resistance

The first antibiotic discovered with AI is halicin. There are many instances of using AI partly to identify antibiotics. But this new antibiotic is the ultimate result of fully being dependant on AI without any human assumptions previously. This research has been published on Cell. The leader of the research team is Jim Collins, a synthetic biologist from the Massachusetts Institute of Technology in Cambridge.  

According to Jacob Durrant, this research work is remarkable. He is a computational biologist from the University of Pittsburgh, Pennsylvania. He says so because besides identifying candidates, the study finds out and validates promising molecules from the animal tests. Further, other types of drugs like that are targeting to treat cancer, and other diseases can apply this approach.

Day by day, more bacteria are being resistant to antibiotics. So, scientists must invent new antibiotics so that they can stop the killer resistant infections form taking away millions of people by 2050. Although the need is urgent, we have seen a slow growth rate of discovering and approving new antibiotics in the last few decades. Collins says, “People keep finding the same molecules over and over. We need novel chemistries with novel mechanisms of action.”  

No More Assumptions

The research team, led by Collins used the brain’s architecture to build an AI algorithm and finally formed a neural network. As the algorithm is flawless, it studies the molecules’ properties atom by atom.

Then the neural network got the training to spot molecules by the researchers. Escherichia coli is a harmful bacterium. The spotted molecules can inhibit the growth of that bacteria. They had a collection of 2,335 units with known antibacterial activity. This includes a library of about 300 approved antibiotics, as well as 800 natural products from the plant, animal, and microbial sources.

According to Regina Barzilay, an MIT AI researcher, the algorithm does not depend on assumptions like drugs’ activity and labeling chemical groups. She says, “As a result, the model can learn new patterns unknown to human experts.”  

After the model receives training, researchers use the Drug Repurposing Hub, a library for screening the model. The hub is capable of handling investigations on 6,000 molecules of human diseases. The team commanded the model to predict effective processes against E. coli and also separate molecules that are different in their appearance. 

Bacteria’s Are Becoming More Resistant Than We Thought

For further testing, the team selected a group of 100 candidates to run tests on them. Among the molecules, one such was used as a diabetes treatment. It was a potent antibiotic, according to the team, and named this as halicin. They named it after HAL, the intelligent computer from the movie 2001: A Space Odyssey. Being tested on the mice, it showed active activities against many harmful pathogens.

Proton Block

The working process of antibiotics combines some mechanisms. For example, we can take enzyme blocking in protein synthesis, cell-wall biosynthesis, and DNA repair. But halicin antibiotic is unique with its mechanism as it does not follow the conventional proton flow across the cell membrane. Experts noticed it has low toxicity and fights resistance. Collins says, “But even after 30 days of such testing, we didn’t see any resistance against halicin.”  

In the next step, the team started screening 107 million molecular structures. The database containing massive data is known as ZINC15. The shortlisted 23 among them and 8 of them have antibacterial activity. Two of the eight molecular structures can fight a lot of pathogens and even overcome antibiotic-resistant strains of E. coli.

Another computational biologist from Carnegie Mellon University in Pittsburgh, Bob Murphy, says, “The study is a great example of the growing body of work using computational methods to discover and predict properties of potential drugs.” He believes the bulk analysis was possible as they used AI to analyze the enormous database.

Collins and the team do not want to use this approach as searching for specific structures. They want to train the network to identify molecules with a particular activity. Researchers are now waiting for an outside company or group that will continue halicin’s clinical trials. Further research can open the way to find more new antibiotics. Barzilay says, “This study puts it all together and demonstrates what it can do.”