For the first time in more than six decades, scientists have identified a new class of antibiotics — and they used artificial intelligence to do it. A team of 21 researchers at the Massachusetts Institute of Technology trained deep-learning models to screen millions of chemical compounds, zeroing in on candidates capable of killing methicillin-resistant Staphylococcus aureus, the drug-resistant superbug known as MRSA. The results were published in Nature in December 2023 C.E.
At a glance
- MRSA antibiotic discovery: Two promising antibiotic candidates from the same chemical class were identified after screening roughly 12 million commercially available compounds using deep-learning models.
- Deep learning models: Researchers trained additional models to assess toxicity across three types of human cells, allowing the team to separate compounds that kill bacteria from those that also harm the body.
- Mouse model results: In lab experiments testing both skin and systemic MRSA infection, each candidate compound reduced the MRSA bacterial population by a factor of 10.
Why MRSA makes this matter so much
MRSA is not a single disease. It is a shape-shifter — capable of causing mild skin infections in one patient and life-threatening pneumonia or bloodstream infections in another. Nearly 150,000 MRSA infections occur each year in the European Union alone. Across the E.U., almost 35,000 people die annually from antimicrobial-resistant infections, according to the European Centre for Disease Prevention and Control.
Globally, the picture is just as grim. The World Health Organization has listed antimicrobial resistance as one of the greatest threats to human health. Decades of antibiotic overuse have accelerated the problem, while the pipeline of genuinely new antibiotic classes has been dry for more than 60 years. That gap is what makes this discovery so significant.
Opening the black box
Earlier AI drug discovery efforts drew criticism for being opaque — models that produced results without anyone fully understanding why. This team deliberately changed that. “What we set out to do in this study was to open the black box,” said Felix Wong, a postdoc at MIT and Harvard and one of the study’s lead authors.
The MIT team trained an expanded deep-learning model on data from approximately 39,000 compounds evaluated for antibiotic activity against MRSA. Chemical structure data was fed in alongside the activity results, allowing the model to learn which molecular features predicted effectiveness.
Professor James Collins, who holds the Thomas D. and Virginia W. Cabot Professorship in Medical Engineering and Science at MIT, described the advance in practical terms. “Our work provides a framework that is time-efficient, resource-efficient, and mechanistically insightful, from a chemical-structure standpoint, in ways that we haven’t had to date,” he said.
Three additional models then evaluated toxicity on human cells. By combining antimicrobial effectiveness with low human toxicity, the team narrowed a field of 12 million compounds down to around 280 worth testing in a lab. From that set, two strong candidates emerged — both from the same newly identified chemical class.
What comes next
Moving a compound from mouse models to human clinical trials is a long, uncertain road. Neither candidate has yet been tested in humans, and many promising antibiotic leads have failed at later stages of development. The pharmaceutical industry’s limited financial incentives for antibiotic development remain a structural problem that this discovery alone cannot fix.
Still, the method itself may be the bigger story. The transparent deep-learning framework the MIT team built is not limited to MRSA — it could be applied to other drug-resistant pathogens, accelerating the search for treatments against bacteria that currently have no good answers. At a moment when the antibiotic pipeline has been described as critically depleted, a reproducible, efficient way to find new classes of compounds is exactly the kind of structural change the field needs.
Sixty years is a long time to wait. This research suggests the wait for the next discovery may be considerably shorter.
Read more
For more on this story, see: Euronews
For more from Good News for Humankind, see:
- Renewables now make up at least 49% of global power capacity
- Alzheimer’s risk cut in half by drug in landmark prevention trial
- The Good News for Humankind archive on global health
About this article
- 🤖 This article is AI-generated, based on a framework created by Peter Schulte.
- 🌍 It aims to be inspirational but clear-eyed, accurate, and evidence-based, and grounded in care for the Earth, peace and belonging for all, and human evolution.
- 💬 Leave your notes and suggestions in the comments below — I will do my best to review and implement where appropriate.
- ✉️ One verified piece of good news, one insight from Antihero Project, every weekday morning. Subscribe free.
