Using generative AI to discover new antibiotics
On 14 August 2025, the Massachusetts Institute of Technology (MIT) reported1 that a team of its researchers has used generative AI to help design two new antibiotics for treating gonorrhoea and MRSA (Methicillin-Resistant Staphylococcus Aureus). The MIT team led a project where generative AI was given the chemical structure of known compounds alongside other data (including their known antimicrobial properties), and learnt how various bacteria are affected by different molecular structures. The AI was then used to design millions of potential compounds (including compounds that do not currently exist, or have not yet been discovered) and screen them to find the most promising ones, filtering out candidates along the way (e.g. where predicted to be toxic to humans). The project resulted in two different novel compounds that are structurally distinct from any existing antibiotics, which potentially could kill drug-resistant gonorrhoea and MRSA.
The slow-down in the antibiotics pipeline
It is still some way off before these two compounds will be ready for testing in humans in clinical trials. Whilst there is no guarantee that these experimental drugs will be proved safe and effective (and therefore whether they can eventually be prescribed to patients), this does show that generative AI has the potential to be used to design completely new antibiotics. Since the 1970s, the antibiotic pipeline has slowed down dramatically, and many of the more recently approved treatments have been modifications of existing ones. These variants are often less effective, and bacteria can usually develop resistance to them more quickly, in contrast to new classes of antibiotics. However, very few entirely new classes of antibiotics have been developed and approved over the past 50 years or so. New classes of molecules that work in a different way, such as the ones discovered by MIT, are therefore much in demand, and could help combat the global crisis of antimicrobial resistance (AMR).
Could this be the start of a new era in antibiotics drug development?
So, could AI revolutionise antibiotic drug discovery, and spur on a new wave of development? The enormous scientific challenges involved in finding new antibiotics have certainly been a factor in the decline in the drug pipeline over the years. AI promises to be an effective tool in helping to overcome some of these difficulties more quickly and cost-effectively. However, this is only part of the story. Amongst other things, there are other very significant economic challenges when it comes to commercialising antibiotics, which have meant that many drug companies have lost interest in developing new ones.
The economic challenges
Drug development is expensive and risky at the best of times, but with antibiotics, the return on investment can be particularly poor. Antibiotics have to be used sparingly in order to remain effective, and the use of a new antibiotic is generally discouraged, unless cheaper, readily available, generic antibiotics have already been tried by a patient and failed to clear up the infection. Also, they are used over a short duration (e.g. over a few days, rather than months or years). Therefore, pharmaceutical companies are generally more reluctant to undergo time consuming, expensive research or invest in risky products with little market potential.
Incentives
As the traditional sales-based model often fails to cover the high R&D costs and investment needed for a new antibiotic, new business models and innovative pricing strategies are needed for these treatments, in order for this situation to change. With this in mind, some governments are designing incentives to provide a stable revenue stream and a higher potential return on investment, to encourage companies to develop new antibiotics. In particular, some governments have introduced (or are considering) subscription models, where healthcare providers pay a fixed annual fee for access to new antibiotics (for further details, see below). Development of new antibiotics is also encouraged through direct funding and grants for early-stage research, such as drug discovery, preclinical studies and initial clinical trials (known as 'push incentives'). Other incentives can include market entry rewards for successful drug approval, development and commercialisation, or transferable exclusivity vouchers (known as 'pull incentives'). In the latter case, these vouchers would allow a pharmaceutical company to extend the market exclusivity for a different, more profitable, drug in its portfolio.
In the UK, for example, under the Antimicrobial Products Subscription (also known as “Netflix for antimicrobials”), NHS England pays pharmaceutical companies a fixed yearly fee to ensure access to a supply of new antibiotics. This approach provides drug companies investing in antibiotics with a guaranteed income, while also discouraging the overuse of these medicines. A similar subscription model approach has been proposed in the US with the PASTEUR (Pioneering Antimicrobial Subscriptions To End Upsurging Resistance) Act, where instead of paying for antibiotics based on the volume of drugs sold, the US government would pay a fixed, upfront fee for access to a supply of new, critically needed antimicrobials. Despite this not being law yet, the latest bill introduced in 2023 has the support of a broad coalition of organisations that believe this is a necessary tool to combat the growing rise of antimicrobial resistance and to ensure that the US has an adequate supply of effective treatments to fight drug-resistant infections. The Council of the European Union has also recommended the development of a package of ‘pull incentives’ across the EU, in order to encourage pharmaceutical companies to produce new antimicrobials and ensure access to these drugs in the EU. These might include payment models that have “revenue guarantees” for pharmaceutical companies, similar to those offered by subscription payment models.
Impact on licensing agreements
Academic bodies or research institutes (such as MIT) would usually license out the rights in relation to a discovered drug to a biotech or pharmaceutical company, to further develop and take to market. Most licence agreements for drug compounds are typically structured with the licensee company paying an upfront fee, milestone payments, and royalties on product sales. In the case of antibiotics, those payments would typically be lower than for many other drugs – otherwise the deal risks being commercially unviable for the licensee. For antibiotics, milestone payments tied to sales volumes or sale revenue are generally less suitable, in comparison to those linked to clinical and regulatory milestones, such as the grant of a marketing authorisation. With the introduction of new subscription based models, payment guarantees and similar schemes, as well as other incentives, for antibiotics, the negotiation power of licensors may improve with time. However, as these initiatives are still in their early stages, it remains to be seen whether they will have a real impact on the terms or dynamics of these deals in the future.
Closing remarks
In summary, it is easy to predict the important and disruptive role that AI can play in the discovery of antibiotics. We have already seen how AI can significantly reduce the time and cost of developing these much needed drugs (not just in the discovery stage but also, for example, by streamlining and optimising clinical trials). Nevertheless, other initiatives and incentives are called for in order to make it commercially viable for companies to produce and sell antibiotics, and therefore invest in their development in the first place. However, this news is encouraging and undoubtedly is a step in the right direction.
Footnotes:
1See the report from MIT (here), plus other reports in the press such as the BBC (here).