Deep learning AI Discovered New Antibiotic for the Very First Time

Humans have been using antibiotics for about 100 years. For 30 years the competent authorities have been warning of the future problem of antibiotic resistance, since higher doses or different antibiotics are needed to end them. Pathogens, whether bacteria, fungi or protists, that have been traditionally stopped with antibiotics have naturally developed resistance to the drugs used against them. This is due to the process of constant evolution that occurs in nature, but health authorities point out that the misuse and abuse of antibiotics has helped this adaptation to take place much faster than expected. Health authorities suggest that by the end of the 21st century the current antibiotics will no longer be useful and that from 2050 we may already notice the lack of response from many of them. It is estimated that by then about 10 million people will die each year from resistant infections.

We are losing the antibiotics that defense us of the pathogens

Antibiotics kill or prevent the growth of pathogens that invade the body. However, for this to have a full effect the treatments must be prolonged and in the recommended amounts. If these two conditions are not met, we will kill a part of the bacterial population. If, when the individual begins to feel better, the recommended treatment is stopped early, it is very possible that the person will recover from the infection. However, those pathogens that due to sheer genetic variability were slightly more resistant to the antibiotic will survive. In this scenario, these pathogens will have been selected to survive and will leave their offspring with that trait that slightly protects them. When this happens repeatedly we will eventually have a population with increasing resistance. In addition, to this must be added the ability of some bacteria to transmit genetic material horizontally, to other individuals without the need for reproduction or genetic recombination.

The search for new antibiotics has been a constant race for half a century. While it is true that new promises in the field are discovered with extensive research and development efforts every few years, the speed with which these molecules are found is less and less. It is estimated that one is found every 15 years at a cost of about 1,190 million dollars. Some of the most promising antibiotics that have recently started to be developed have a particularity, they have been generated by artificial intelligence (AI). It is true that predictive software was already used at various points in the process, but this is the first time that an artificial intelligence has proposed a new chemical compound. The artificial intelligence started from a database with more than 2,300 compounds with known antibacterial properties and of both plant, animal or bacterial origin. Among them were 300 that were already authorized antibiotics. Thanks to this, the AI learned what kinds of behaviors were being looked for and to predict the molecular function of a compound when it was presented with the set of proteins of the most common pathogens. When the machine had learned, 6,000 compounds were proposed to it that are currently being used in different medical research and that were not the known antibiotics and their possible chemical variants. The AI established an order from higher to lower probability of success as antibiotics of the different compounds with respect to the E. coli bacteria.

The AI could save time for the antibiotic discovery race

But the artificial wonder does not end here. In a new phase of testing computing capabilities, a database of more than 100 million chemical molecules was given the AI. In just 48 hours, more than two dozen molecules had already been selected with very good probabilities. A laboratory team was tasked with testing the efficacy of these compounds. The results showed that 50% of them did indeed have antibiotic activity. This might seem like a very low number of them, but you have to remember that they are all molecules that we already know and have never tried as antibiotics. Right now, and thanks to AI, a huge amount of resources have been saved dedicated to determining which molecules should be studied. The most promising of all those that were selected has been a drug that is currently being developed to treat diabetes. The molecule was christened Halicin, in honor of the ship's on-board computer from "2001: A Space Odyssey" HAL. Laboratory results in mouse experiments of halicin are very encouraging. It has shown its effect both in strains of E. coli and in some of the antibiotic resistant superbugs such as some strains of Acinetobacter baumannii. Furthermore, the tests show low toxicity to mice and no resistance in treatments of up to 30 days.

If this is so, perhaps we can change the predictions regarding antibiotic resistance during the 21st century and find new antibiotics with which to continue developing new variants. But the implications of the results obtained with deep learning go further. These trials open a door to a whole new way of looking for drugs in other areas such as cancer treatment or rare diseases.


No time to wait: Securing the future from drug-resistant infections. Report to the Secretay-General of the United Nations. World Health Organitation. April 2019.

Munita JM, Arias CA. Mechanisms of Antibiotic Resistance. Microbiol Spectr. 2016 Apr;4(2):10.1128/microbiolspec.VMBF-0016-2015.

Davies J, Davies D. Origins and evolution of antibiotic resistance. Microbiol Mol Biol Rev. 2010 Sep;74(3):417-33

Jonathan M. Stokes, Kevin Yang, Kyle Swanson, Wengong Jin, Andres Cubillos-Ruiz, Nina M. Donghia, Craig R. MacNair, Shawn French, Lindsey A. Carfrae, Zohar Bloom-Ackermann, Victoria M. Tran, Anush Chiappino-Pepe, Ahmed H. Badran, Ian W. Andrews, Emma J. Chory, George M. Church, Eric D. Brown, Tommi S. Jaakkola, Regina Barzilay, James J. Collins. A Deep Learning Approach to Antibiotic Discovery. Cell, Volume 180, Issue 4, 2020, Pages 688-702.e13.


About Us · User Accounts and Benefits · Privacy Policy · Management Center · FAQs
© 2021 MolecularCloud