Technology at the Forefront in Tackling Antimicrobial Resistance: A Conversation with Dr. Jim Collins, Termeer Professor of Medical Engineering and Science at MIT

With antibiotic resistance looming as one of the greatest public health threats of our time, Dr. Jim Collins, the Termeer Professor of Medical Engineering and Science at MIT, has once again broken new ground. Dr. Collins is a pioneer in synthetic biology and in applying artificial intelligence to antibiotic discovery, bringing a fresh, high-tech perspective to a critical challenge. His recent research, published in Nature, demonstrates how AI can unlock new ways to discover antibiotics—a groundbreaking step toward combating drug-resistant bacteria. In this conversation, Dr. Collins shares his insights on the potential of AI, the role of synthetic biology, and the broader vision he sees for a future where AMR is more effectively managed.

Dr. Collins, thank you for joining us. Your recent Nature publication has generated significant attention for its innovative application of AI in antibiotic discovery. Could you tell us how this approach came about?

Jim Collins: We started tackling this problem about five years ago through a collaboration at MIT with my colleagues Regina Barzilay and Tommi Jaakkola. We wanted to see how deep learning could be applied to antibiotic discovery, and in our initial study, we took a straightforward approach. We compiled a training set of about 2,500 compounds and tested them on E. coli to identify those with antibacterial activity, showing at least 80% growth inhibition.

We then trained a graph neural network—a deep learning model that examines chemical structures bond-by-bond and substructure-by-substructure—to classify compounds as antibacterial or not. The model learned from our training set and was then able to evaluate new compounds from virtual libraries, predicting their potential as antibiotics based purely on chemical structure. Our tests showed a true positive rate of about 52%, which is impressive compared to random screening. We’ve since expanded this model to work with much larger libraries and additional pathogens, giving us a powerful AI-based tool for antibiotic discovery.

What inspired you to take this AI-based approach to antibiotic discovery?

Jim Collins: I believe we’re at the cusp of a paradigm shift in antibiotic discovery. Traditional methods for finding new antibiotics have been slow and increasingly inefficient. AI has the potential to significantly accelerate the discovery process and cut costs. Traditionally, drug screening—even for antibiotics—involves gathering a massive screening library that might include 10,000, 100,000, a million, or even more compounds. These compounds are then screened experimentally against a target pathogen, but it’s time-consuming to curate such large libraries and conduct the screenings.

With AI, though, we can train models on much smaller libraries and use those models to explore vast chemical spaces. For example, we started with libraries trained on 2,500 compounds and have since scaled up to 39,000. This has allowed us to screen in silico libraries containing billions of compounds—libraries that would be impossible to handle experimentally. With AI, we can now screen these virtual libraries in a matter of days, exploring larger chemical spaces with far greater efficiency and at a lower cost.

You are a pioneer in synthetic biology and systems biology. How have these fields shaped your approach to addressing AMR?

Jim Collins: It’s been absolutely crucial. AMR is a complex issue that spans microbiology, genetics, pharmacology, epidemiology, and now, AI. Almost 25 years ago, we helped launch synthetic biology, realizing we could apply engineering principles to molecular biology. This led us to design biological circuits made up of interacting DNA, RNA, and protein components. Since then, we’ve used synthetic biology to address specific AMR challenges.

For instance, we developed RNA switches that let us manipulate genes within a bacterial pathogen. This approach allows us to reconstruct the underlying network and better understand how antibiotics work. We also created inducible switches to study the biological effects of essential genes in bacteria—genes that can’t be studied through knockout because the bacteria won’t grow without them. With these systems, we can fine-tune gene expression to determine if a gene would make a good drug target.

Additionally, we’ve engineered phages as antibiotic replacements and adjuvants. In one project, we showed that phages could be engineered to overexpress proteins, which can boost antibiotic effects by 1,000 to 100,000-fold, even re-sensitizing resistant strains. Lastly, we pioneered the development of cell-free RNA switches for low-cost diagnostics to identify the bacterial species causing an infection, distinguish it from viral infections, and assess antibiotic susceptibility. We’re now launching a major effort with Community Jameel to develop these low-cost diagnostics for AMR.

Do you see potential for integrating synthetic chemistry with synthetic biology in this work?

Jim Collins: I do, and I’ll give you an example. We have a new project in collaboration with colleagues in Korea and across the U.S. The goal is to biomine the metagenomics space—essentially searching bacterial, plant, fungal, and insect genomes to identify biosynthetic gene clusters that might yield new antimicrobials. Here, AI plays a crucial role. We can use it to identify these gene clusters, predict their outputs, and determine if they might be effective antibiotics.

We can then use a combination of synthetic chemistry and synthetic biology to reconstruct these gene clusters in a production organism and produce the molecules they encode. Additionally, we’re exploring a modular approach, combining synthetic chemistry, synthetic biology, and AI to identify key modules in successful biosynthetic gene clusters for antibiotics. The idea is that we could eventually use these modules to create entirely new molecules that don’t even exist in nature.

How does this progress fit into your broader vision for combating AMR?

Jim Collins: The approaches we’re taking can address AMR in a few important ways. First, we’re focused on discovering and designing antibiotics with novel mechanisms of action. By creating drugs with new mechanisms, we make it harder for resistance developed against existing antibiotics to affect them. Resistance will still arise, but our AI tools help us stay a step ahead, building a pipeline that can deliver new molecules as resistance evolves.

Secondly, I think AI is particularly well-suited to finding molecules that target more than one site. This lets us use polypharmacology as an advantage, rather than a “bug” to optimize out. With multiple targets in the mix, it becomes harder for resistance to develop, which could extend the lifespan of these drugs significantly.

More broadly, one of the big gaps in drug discovery is predicting interactions between small molecules and proteins. Our prediction capabilities aren’t where they need to be. I’m hopeful that AI will advance much faster and more accurate methods for molecular docking, which could transform antibiotic discovery and drug development in many fields.

Beyond science and technological development, what do you think are the key steps needed to translate these discoveries into real-world solutions for AMR?

Jim Collins: Partnerships are essential for translating academic discoveries into real-world solutions. Many of the technologies we develop in the lab have potential in the fight against AMR, but they need scaling up and commercialization to reach broad use. One example is our collaboration with Phare Bio, a nonprofit we launched as part of the Antibiotics-AI Project. Phare Bio is advancing our AI-discovered antibiotic candidates, doing the detailed analysis and development work needed to move them toward clinical use. It’s a powerful partnership, and I believe it will be crucial in developing the next generation of tools to combat AMR.

Another big gap is funding for later-stage trials. We need more resources here. Market and social incentives are also important to encourage governments, biotech, and pharma to re-engage in this space. AI can help by reducing early discovery costs and boosting the likelihood of success in clinical trials, but sustainable financing will be critical for moving candidates through the pipeline.

Thank you for these insights, Dr. Collins. Any final thoughts on what lies ahead?

Jim Collins: I think the next decade will bring more resistance, unfortunately, because we’re not fully equipped yet. But I’m optimistic that we’ll also see greater awareness that AMR is one of the major challenges facing humanity. The pace of progress is promising, and I think we’ll see AI playing a growing role in both early discovery and later-stage development.

AI will also help us predict resistance by analyzing both epidemiological data and protein-level interactions. These predictions will allow us to identify whether a drug target is robust enough to withstand mutations or if it’s vulnerable, and AlphaFold and similar projects are likely to give us the tools we need. I’m hopeful that we’re heading toward a second golden age in antibiotic discovery, powered by AI and inspired by the talented young researchers entering this field. They understand that these are exciting problems with real-world impact, and that’s what will drive meaningful change.