German researchers and industry partners have launched a new public-private programme aimed at accelerating the discovery of urgently needed antibiotics using artificial intelligence, high-throughput laboratory testing and explainable machine learning.
The ANTARES partnership brings together the Fraunhofer Institute for Translational Medicine and Pharmacology ITMP in Hamburg, the Helmholtz Centre for Infection Research in Braunschweig and pharmaceutical research company Enamine Deutschland.
Funded by the German Federal Ministry of Research, Technology and Space, the programme will use AI models to identify promising antibacterial compounds more quickly and improve the chances that potential drug candidates progress beyond the early discovery stage.
The initiative responds to the growing global threat of antimicrobial resistance, which is reducing the effectiveness of existing treatments while the commercial market for developing replacement antibiotics remains challenging.
Combining prediction with laboratory validation
ANTARES will initially train and test its AI models using publicly available datasets, allowing researchers to compare different approaches before applying them to newly synthesised chemical libraries.
Rather than relying on a single round of model training and testing, the programme will use an iterative process in which AI predictions are repeatedly assessed through high-throughput experimentation.
Compounds predicted to have antibacterial activity will be tested in the laboratory. Structurally similar compounds predicted to be inactive will also be examined, helping researchers determine whether the model can reliably distinguish between promising and ineffective molecules.
The results will then be fed back into the model to improve its accuracy before the next round of predictions.
Dr Bernhard Ellinger of Fraunhofer ITMP said this repeated cycle of prediction and experimental validation differentiates ANTARES from many other AI-led drug discovery initiatives.
“The iterative rounds of predictions and experimental validations are a big plus of our approach,” he said. “Other programmes often rely on a single round of training the model, predicting actives and validating them experimentally. This is where ANTARES will make a difference.”
The consortium will also have the capability to design, synthesise and test entirely new molecular libraries rather than remaining dependent on existing datasets.
Addressing the antibiotic development gap
The programme is intended to help address the structural weaknesses in the global market for antibiotic development.
New antibiotics are medically essential, but their use is often deliberately restricted to slow the development of resistance. This can limit commercial returns and make investment in research less attractive than in therapeutic areas where medicines are prescribed more frequently.
Dr Sven Wagner, Vice President of Partnerships at Enamine, described the situation as a “broken market” in which urgent medical need is not matched by a viable return on investment for drug developers.
He said public-private partnerships were essential to helping research programmes cross the funding and development gaps that frequently prevent early scientific discoveries from becoming usable treatments.
Within ANTARES, Enamine will support the design and production of small molecules at scale, enabling the consortium to test a wider and more diverse range of potential antibiotic compounds.
Making AI decisions more transparent
Explainable AI will form an important part of the programme.
Researchers will use explainability techniques to identify the structural features that influence a model’s predictions and to better understand why particular molecules may show antibacterial activity.
This is intended to reduce reliance on opaque, black-box outputs and give scientists greater confidence when deciding which compounds should move forward for further testing.
A generative AI model will also be used to propose new molecular structures with potential antibiotic activity.
The resulting models and standardised datasets are expected to be published in line with FAIR principles, meaning they should be findable, accessible, interoperable and reusable by other researchers.
Professor Mark Brönstrup of the Helmholtz Centre for Infection Research said the objective was to reduce discovery timelines and improve success rates by combining computational prediction with experimental evidence.
The programme will also investigate the mechanism of action and range of activity of promising compounds, helping determine how they affect bacteria and which infections they may be capable of treating.
Project management perspective
ANTARES demonstrates how AI-enabled research programmes increasingly depend on tightly integrated delivery models rather than separate scientific workstreams.
The partnership combines data science, medicinal chemistry, high-throughput screening, microbiology and drug development expertise. Each AI prediction must connect with compound synthesis, laboratory testing, data interpretation and subsequent model refinement, making interface management central to the programme’s success.
Its iterative structure closely resembles an Agile delivery model. Instead of waiting until the end of the research programme to validate the technology, the consortium will work through repeated cycles of prediction, testing, learning and adjustment. This should allow weak approaches to be identified early while directing resources towards the most promising models and molecules.
However, iterative delivery does not remove the need for scientific governance. The consortium will need agreed data standards, quality controls, decision thresholds and validation protocols to ensure that results generated by different partners remain comparable and reproducible.
Explainable AI also introduces an important assurance mechanism. In a high-risk field such as drug discovery, model accuracy alone may not be sufficient. Researchers, funders and future regulatory stakeholders will need to understand how predictions were generated and whether they are supported by credible biological and chemical evidence.
For project leaders, ANTARES highlights the importance of structuring AI programmes around measurable scientific outcomes rather than treating the technology as the objective. The value of the initiative will not be determined by the sophistication of its models, but by whether the partnership can identify stronger antibiotic candidates, reduce wasted experimentation and move viable discoveries towards further development.
The programme also shows why public-private collaboration is becoming increasingly important in areas where commercial incentives alone may be insufficient. By combining government funding, public research capability and industrial drug discovery expertise, ANTARES is attempting to create a delivery model capable of progressing medically important innovation through one of the most difficult stages of the pharmaceutical development lifecycle.
