Antimicrobial resistance is one of the defining challenges of modern medicine. Infections that were once easily treated with penicillin or tetracycline are now harder to control, and resistant pathogens such as methicillin-resistant Staphylococcus aureus (MRSA) and drug-resistant Neisseria gonorrhoeae are becoming increasingly common. Global health authorities warn of a possible post-antibiotic era, where routine infections could once again become life-threatening.
In parallel, the pace of traditional antibiotic discovery has slowed. The last few decades have seen few genuinely novel classes of antibiotics, with most new drugs representing only modifications of existing compounds. In this context, artificial intelligence (AI) is emerging as a powerful tool to revitalise antibiotic development. By combining computational methods with advanced medical imaging, researchers can not only discover new molecules but also visualise their effects in ways that were previously impossible.
From screening to algorithms
The so-called golden age of antibiotic discovery in the mid-twentieth century relied heavily on screening natural products from soil microorganisms. This approach yielded classics such as streptomycin, vancomycin and erythromycin. Over time, however, the pipeline dried up: new screens often rediscovered the same compounds, while escalating regulatory costs and limited commercial incentives discouraged pharmaceutical investment.
AI offers an alternative path. Instead of laboriously screening thousands of extracts in the laboratory, algorithms can predict which molecules are likely to have antibacterial activity. This is achieved by training models on large datasets of molecular structures and biological outcomes. Once trained, these systems can explore vast areas of “chemical space” in silico, reducing the number of compounds that need to be tested experimentally.
The power of this approach was first demonstrated in 2020 with the discovery of halicin, an antibiotic identified by MIT researchers using deep learning. Halicin, structurally distinct from known antibiotics, proved active against resistant bacteria, including Clostridioides difficile and Mycobacterium tuberculosis. This landmark result hinted at what could be achieved if AI-driven discovery were scaled up.
Generative AI and the creation of new antibiotics
In 2025, MIT scientists took this further by using generative AI algorithms to design new antibiotics from scratch. Reporting in Cell, the team created more than 36 million theoretical compounds and screened them computationally for antimicrobial activity. From this vast library, they identified two molecules with striking promise.
The first, NG1, was designed to combat drug-resistant N. gonorrhoeae. It works by interfering with LptA, a protein involved in building the bacterial outer membrane. Laboratory and mouse studies showed that NG1 could effectively clear resistant gonorrhoea infections. The second, DN1, emerged from an unconstrained generative design aimed at Gram-positive bacteria. DN1 showed strong activity against MRSA, clearing skin infections in mouse models.
Both compounds are structurally novel, acting through mechanisms that differ from current antibiotics. This illustrates how generative AI can move beyond modifying known molecules, enabling the design of drugs with entirely new biological targets.
Where medical imaging fits in
Although algorithms can design and predict molecular activity, experimental confirmation is essential. This is where medical imaging becomes crucial, linking computational predictions with real biological outcomes.
At the microscopic level, advanced imaging such as cryo-electron microscopy and super-resolution fluorescence microscopy allows scientists to visualise how antibiotic candidates interact with bacterial membranes or intracellular targets. For NG1, imaging confirmed disruption of membrane synthesis pathways in N. gonorrhoeae, consistent with its predicted mechanism of action.
In preclinical and clinical contexts, modalities such as PET (positron emission tomography) and MRI play a role in tracking how antibiotics are distributed within the body. Radiolabelled versions of antibiotic candidates can be followed in real time, revealing whether the drug reaches infected tissues in sufficient concentration. This helps bridge the gap between activity in a laboratory dish and efficacy in a living organism.
For example, PET imaging has been used to study the biodistribution of fluoroquinolones and rifampicin, providing insights into how these drugs penetrate lungs, abscesses or urinary tract tissues. Applying the same approach to AI-generated antibiotics such as NG1 or DN1 will be essential to optimise dosing strategies and to predict therapeutic success.
Imaging also plays a role in measuring treatment response. CT and MRI scans can show the resolution of abscesses or lung infiltrates, while molecular imaging can track the decline of bacterial metabolic activity. These tools allow researchers to quantify not just whether a drug works, but how quickly and in which tissues it is most effective.
AI, imaging and personalised therapy
The integration of AI and imaging opens possibilities for more personalised approaches to infectious disease treatment. Machine learning systems can analyse imaging data alongside genomic and clinical information to predict how individual patients will respond to specific antibiotics. For instance, imaging biomarkers of infection burden could be matched to AI-driven models of drug penetration, helping clinicians select the most effective treatment for each case.
In research, imaging datasets themselves can be used to train AI models. For example, machine learning applied to MRI or CT scans may help quantify infection spread or monitor resistance dynamics over time. Combining these imaging insights with generative AI drug design could create a feedback loop: new antibiotics are designed, tested, imaged in vivo, and the imaging results feed back into refining the AI models.
Challenges and limitations
Despite the excitement, challenges remain. AI models are only as good as the data they are trained on. For generative design, chemical datasets need to be accurate, diverse and well-annotated. Imaging data must be carefully standardised to ensure reproducibility across different centres and platforms.
Another limitation is the complexity of translating predictions into safe, effective medicines. Toxicity, poor absorption or unintended side effects can still derail promising compounds. Imaging can help detect these problems earlier — for example, by revealing unexpected accumulation of drugs in non-target tissues — but clinical validation remains a lengthy and costly process.
Regulators will also demand transparency. Many AI models function as “black boxes”, generating predictions without clear explanations. Linking predictions to imaging evidence of mechanism may help build the case for regulatory approval.
The way forward
Rather than replacing laboratory research, AI and imaging are best viewed as complementary tools. Together, they expand the scope of discovery and deepen the understanding of how candidate antibiotics behave in complex biological systems. The MIT discoveries of NG1 and DN1 show that generative AI can imagine drugs outside the reach of conventional chemistry, while imaging provides the means to verify and refine those discoveries.
Looking ahead, international collaboration will be vital. Building shared databases that combine chemical, genomic, clinical and imaging data will ensure that AI systems are trained on the richest possible information. Investment from governments, non-profits and industry will be needed to push promising molecules through the costly preclinical and clinical stages.
Conclusion
Antibiotic resistance is a mounting crisis, but the convergence of artificial intelligence and medical imaging offers new hope. Generative algorithms can explore chemical spaces beyond human imagination, while imaging technologies provide direct evidence of how those molecules act in living systems. Together, they have the potential not only to discover new antibiotics such as NG1 and DN1, but also to accelerate their journey from digital prediction to clinical reality.
If these tools are fully integrated, the next generation of antibiotics may be discovered and validated faster than ever before. In doing so, AI and imaging could help secure a future where infectious diseases remain treatable, and the post-antibiotic era can be averted.
Disclaimer
The information presented in Artificial Intelligence, Antibiotic Discovery, and the Role of Medical Imaging is intended for educational and informational purposes only. It does not constitute medical, scientific, or regulatory advice. While every effort has been made to ensure accuracy, the field of antibiotic discovery and artificial intelligence is rapidly evolving, and details may change as new research emerges.
References to specific compounds, such as NG1, DN1, or halicin, are based on published preclinical or experimental studies. These compounds are not currently approved for clinical use, and their safety and efficacy in humans remain under investigation. Readers should not interpret this article as evidence of clinical effectiveness or as a substitute for professional medical consultation.
Any mention of medical imaging techniques is illustrative and should not be considered guidance for diagnosis or treatment. Decisions regarding patient care, drug development, or therapeutic strategies must be made in consultation with qualified healthcare professionals and regulatory authorities.
The authors and publishers accept no responsibility for any loss, injury, or damages arising from reliance on the information provided in this article.
You are here: home »