AI is reviving the antimicrobial pipeline – just in time
– by Akhila Kosaraju

The views and opinions expressed in this article are solely those of the original author(s) and do not necessarily represent those of GARDP, their donors and partners, or other collaborators and contributors. GARDP is not responsible for the content of external sites.

Most of us take antibiotics without a second thought. We swallow a pill, the infection clears, and happily, we move on. That quiet reliability is one of the greatest achievements of modern medicine.

But that reliability is quickly eroding. Antibiotics are losing effectiveness faster than we are replacing them.

In 2021 alone, antibiotic resistance contributed to about 4.7 million deaths worldwide, including roughly 1.1 million deaths directly caused by resistant infections.¹ Without meaningful action, The Lancet projects that drug-resistant infections could be associated with more than 8 million deaths per year by 2050.¹

In addition to the well-recognized market challenges that limit investment in antibiotics, this crisis persists in part because too few promising discoveries make it far enough along the development path to reach patients.

At the earliest stages of antibiotic research, the probability that a compound will become an approved drug is vanishingly small, on the order of one to two successes out of 10,000 initial compounds.² By contrast, once a compound reaches the end of preclinical development and enters Phase I clinical trials, its probability of eventual Food and Drug Administration (FDA) approval improves to about one in ten.²

Rather than relying on slow, trial-and-error screening, AI enables scientists to more efficiently explore vast, previously inaccessible chemical space and identify promising compounds that might otherwise remain hidden.

Using AI to bridge the gap

This phase between discovery and clinical trials, often called the valley of death, is where scientific risk is highest and capital is scarcest. It is also where the greatest opportunity exists to improve outcomes by generating more hits with artificial intelligence (AI) and being more selective about which compounds are advanced into downstream development – taking fewer but better shots on goal.

Increasingly, researchers across academia, biotech, and public-interest organizations are turning to AI to help bridge this gap. Rather than relying on slow, trial-and-error screening, AI enables scientists to more efficiently explore vast, previously inaccessible chemical space and identify promising compounds that might otherwise remain hidden.

But discovery is only the first hurdle. AI is also being used to help promising molecules survive the difficult transition from an exciting finding in the lab to a viable drug candidate.

While discovery-focused models search large chemical and biological datasets to identify new antibacterial compounds, a second set of AI tools focuses on translation – predicting how these compounds will behave long before they reach the clinic. By modeling properties such as toxicity, pharmacokinetics, and resistance potential, researchers can identify potential liabilities earlier and prioritize the candidates most likely to advance toward clinical trials.

These models can evaluate multiple properties in parallel, but their performance is ultimately constrained by the availability of high-quality, biologically relevant training data – something that remains sparse for many of the properties required of a successful clinical antibiotic. As a result, generating new data alongside model development is becoming an essential part of making these approaches truly predictive.

Even with these constraints, early breakthroughs have demonstrated the potential of these approaches in practice.

Reviving the pipeline with AI breakthroughs

One of the most visible inflection points came in 2020, when researchers led by Jim Collins at MIT identified the first-ever novel antibiotic candidate using AI – repurposing a previously studied compound by uncovering its unexpected antibacterial properties.³ The model learned which chemical features were associated with activity against specific pathogens, and then screened large chemical libraries in silico to uncover compounds that had been overlooked by traditional methods. One of these molecules showed potent activity against Acinetobacter baumannii, a pathogen the World Health Organization has classified as a critical priority.⁴

Since that breakthrough, the field has moved beyond predictive screening toward generative approaches that design entirely new molecules from scratch.

In addition to evaluating existing compounds, generative AI models can create novel structures while balancing multiple design constraints – such as potency, safety, and oral bioavailability – from the outset.

Applied in practice, these methods have already led to the identification of multiple additional novel antibiotic classes, with candidates demonstrating improved efficacy and lower predicted toxicity compared with those emerging from earlier, predictive approaches.^5-7

This shift is also reshaping how scientists approach antimicrobial peptides, a promising but historically difficult class of antibiotics.

At the University of Pennsylvania, César de la Fuente and his collaborators have used AI to discover thousands of previously unknown antimicrobial peptides embedded within the human proteome and across evolutionary datasets.^8-11 By training models to identify peptide sequences with antibacterial activity, the team has dramatically accelerated discovery while uncovering candidates with novel mechanisms of action.

AI’s impact is extending beyond small molecules and peptides to the broader microbial ecosystem as well. A recent Nature Biotechnology article highlighted how machine learning is being applied to microbiome data to better understand microbial interactions, predict therapeutic responses, and identify new intervention strategies.¹²

AI alone cannot resolve the structural challenges facing antibiotic development, nor will it solve the antibiotic crisis without parallel advances in clinical progress, economic incentives, and stewardship.

These approaches are helping researchers move from descriptive microbiome studies toward actionable, translational insights that could inform both therapeutics and diagnostics.

Diagnostics, in particular, represent another critical front. Rapid identification of drug-resistant infections remains a bottleneck in clinical care, often forcing physicians to prescribe broad-spectrum antibiotics while awaiting lab results. AI-enabled diagnostics – using genomics, transcriptomics, imaging, and electronic health record data – are beginning to shorten this window, enabling earlier, more precise treatment decisions and helping preserve the effectiveness of existing antibiotics.¹³

Together, these advances point to a broader shift in how the antibiotic pipeline is being rebuilt – from discovery through translation and diagnosis.

We’re at a pivotal moment where several forces are converging: worsening resistance trends, the extraordinary promise of AI, and a growing recognition that we need new tools to build the next generation of antimicrobial solutions.

Building on AI-driven momentum

AI alone cannot resolve the structural challenges facing antibiotic development, nor will it solve the antibiotic crisis without parallel advances in clinical progress, economic incentives, and stewardship. But by making early-stage research more disciplined, data-rich, and predictive, it allows limited resources to be deployed more effectively – advancing stronger candidates and improving the odds that discoveries survive long enough to benefit patients who urgently need options.

We’re at a pivotal moment where several forces are converging: worsening resistance trends, the extraordinary promise of AI, and a growing recognition that we need new tools to build the next generation of antimicrobial solutions. If we meet this moment with the seriousness it demands, we will look back on this era not as the point when we ran out of options, but when we finally built new ones.

References

1. GBD 2021 Antimicrobial Resistance Collaborators (2024) Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050. 404(10459):1199–1226
2. Boyd NK, Teng C, Frei CR (2021) Brief overview of approaches and challenges in new antibiotic development: a focus on drug repurposing. Front Cell Infect Microbiol.
3. Stokes JM, Yang K, Swanson K, Jin W, Cubillos‑Ruiz A, Donghia NM, et al. (2020) A deep learning approach to antibiotic discovery. 180(4):688–702.e13.
4. World Health Organization (2024) Antimicrobial resistance surveillance and analysis report. WHO. Accessed 7 May 2026
5. Liu G, Catacutan DB, Rathod K, Swanson K, Jin W, Mohammed JC, et al. (2023) Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii. Nat Chem Biol. 19:1342–1350.
6. Wong F, Zheng EJ, Valeri JA, Donghia NM, Anahtar MN, Omori S, et al. (2024) Discovery of a structural class of antibiotics with explainable deep learning. 626:177–185.
7. Krishnan A, Anahtar M, Valeri J (2025) A generative deep learning approach to de novo antibiotic design. 188:5962–5979.e22.
8. Szymczak P, Zarzecki W, Wang J, Duan Y, Wang J, Coelho LP, et al. (2025) AI-driven antimicrobial peptide discovery: mining and generation. Acc Chem Res. 58(12):1831–1846.
9. Torres MDT, Brooks EF, Cesaro A, Sberro H, Gill MO, Nicolaou C, et al. (2024) Mining human microbiomes reveals an untapped source of peptide antibiotics. 187:5453–5467.e15.
10. Santos‑Júnior CD, Torres MDT, Duan Y, Rodríguez del Río A, Schmidt TSB, Chong H, et al. (2024) Discovery of antimicrobial peptides in the global microbiome with machine learning. 187:3761–3778.e16.
11. Maasch JRMA, Torres MDT, Melo MCR, de la Fuente‑Nunez C (2023) Molecular de-extinction of ancient antimicrobial peptides enabled by machine learning. Cell Host Microbe. 31:1260–1274.e6.
12. Nature Editorial (2026) Culturing microbiome therapeutics with big data. Nat Biotechnol. 44:161–162.
13. Arnold A, McLellan S, Stokes JM (2025) How AI can help us beat antimicrobial resistance. npj Antimicrob Resist. 3:18.