Is Machine Learning the Future of Antibiotic Discovery?


Developing novel antibiotics to combat the challenge of increasing antibiotic resistance is harder than it seems. Searching for new compounds is limited by scale and by reliance on a small number of known antibacterial mechanisms. In a new Nature Chemical Biology paper, Liu developed a machine learning model to identify molecules with antibacterial activity against Acinetobacter baumannii, a deadly multi-drug-resistant hospital-acquired bacteria, and discovered a highly specific antibiotic they named abaucin.

Limitations of Current Antibiotics

Antibiotic resistance is one of the most pressing medical challenges we face today. Antibiotic drug resistance occurs through the rapid evolution of bacteria as they experience the selective pressure of antibiotic treatment, with some bacteria mutating and acquiring genetic resistance elements that can then be passed to other surviving bacteria.

When first-line antibiotics fail, there are limited second- and third-line options, and unfortunately many multi- or pan-drug-resistant strains of bacteria have developed that are now untreatable with current antibiotics. The discovery of new antibiotics has been hampered by numerous challenges, and historically has been focused on searching for broad-spectrum agents that target features common to many strains of bacteria. There is a pressing need for novel classes of antibiotics with new mechanisms of action and increased specificity for particular bacteria.

AI for Precision Antibiotic Discovery

One of the limitations of searching for novel antibiotics is scale. Traditional high-throughput screening methods can test a few million molecules for predicted antibacterial activity at their peak scale. Contemporary machine learning methods, on the other hand, can assess hundreds of millions to billions of molecules for potential antibacterial activity in silico, narrowing down a list of promising targets to be validated in the lab.

In a recent paper published in Nature Chemical Biology, Liu et. al. screened roughly 7,500 molecules for their ability to inhibit the growth of Acinetobacter baumannii, a deadly hospital-acquired bacteria that often displays multi-drug resistance. The authors used the screening data to train a message-passing neural network to evaluate molecules for novel antibacterial activity against A. baumannii. As a proof-of-concept, the model was tested on the chemical library from the Drug Repurposing Hub, comprising 6,680 molecules with a variety of structural and chemical properties.

A New Antibiotic Candidate for A. baumannii Infection

After reviewing the model output, the authors found 240 molecules that met their criteria, and tested them for growth inhibition of A. baumannii. Nine molecules exhibited >80% growth inhibition, and after filtering out molecules with similar features to known antibiotics, one molecule emerged as the most promising candidate- RS102895, renamed abaucin for its activity against A. baumannii. 

Notably, abaucin exhibited very specific, narrow-spectrum antibacterial activity against A. baumannii, exerting no effect on other common multi-drug resistant bacteria that were tested. Remarkably, abaucin was highly effective against 41 different strains of A. baumannii, overcoming all of the intrinsic and acquired resistance features of the bacteria across strains.

There are several benefits to the development of narrow-spectrum antibiotics like abaucin. The specificity of abaucin for its target bacteria means there is a lower chance of pre-existing resistance mechanisms or the development of resistance, as the selective pressure of treatment is applied only to one species rather than universally to bacteria in the body. This specificity would likely also reduce dysbiosis, when beneficial bacteria are also killed during antibiotic treatment, resulting in many of the unpleasant side effects associated with antibiotics. This study not only validates the use of machine learning models for discovery of new antibiotics, but produces a promising candidate for treatment of a deadly hospital-acquired infection while helping solve the major biomedical challenge of antibiotic resistance.

Outsourcing Bioinformatics Analysis: How Bridge Informatics Can Help

Groundbreaking studies like these are made possible by technological advances making biological data generation, storage and analysis faster and more accessible than ever before. From building predictive models (machine learning), to pipeline development and software engineering, to deploying existing bioinformatics tools, Bridge Informatics can help you on every step of your research journey.

As experts across data types from leading sequencing platforms, we can help you tackle the challenging computational tasks of storing, analyzing and interpreting genomic and transcriptomic data. Bridge Informatics’ bioinformaticians are trained bench biologists, so they understand the nuances of assay design and the biological questions driving your computational analysis. Click here to schedule a free introductory call with a member of our team.

Jane Cook, Biochemist & Content Writer, Bridge Informatics

Jane Cook, leading Content Writer for Bridge Informatics, has written over 100 articles on the latest topics and trends for the bioinformatics community. Jane’s broad and deep interdisciplinary molecular biology experience spans developing biochemistry assays to genomics. Prior to joining Bridge, Jane held research assistant roles in biochemistry research labs across a variety of therapeutic areas. While obtaining her B.A. in Biochemistry from Trinity College in Dublin, Ireland, Jane also studied journalism at New York University’s Arthur L. Carter Journalism Institute. As a native Texan, she embraces any challenge that comes her way. Jane hails from Dallas but returns to Ireland any and every chance she gets. If you’re interested in reaching out, please email or

Recent Posts