From searching chemical space to designing antibiotics

                                                                                Image designed by Tim Sandle
 

Aarti Krishnan and colleagues focus further on how the rise of artificial intelligence (AI) is transforming the discovery and design of antibiotics. In their recent Viewpoint, “From code to cure: Harnessing AI for antibiotic discovery and design”, published in Science Translational Medicine, the authors examine developments involving both small-molecule antibiotics and antimicrobial peptides, while stressing that computational advances must ultimately translate into medicines that reach patients.[1] 

In the latest Pharmaceutical Microbiology news... 

Traditional antibiotic discovery faces an enormous search problem. The possible chemical space for small molecules is vast, while antimicrobial peptides introduce similarly extensive sequence possibilities. AI offers a means of exploring these spaces more systematically. Krishnan, Hanqun Cao, James J. Collins and Cesar de la Fuente-Nunez describe how AI can support the exploration of chemical and sequence spaces, the design of synthesizable antibiotics and the prioritisation of candidates according to characteristics that include safety and mechanistic diversity.[1]

This represents an important change in emphasis. AI is no longer confined to searching existing chemical libraries for promising antibacterial compounds. Generative approaches create the possibility of designing new antibiotics, potentially extending discovery beyond the chemical structures that have traditionally dominated antibacterial development.

There is already experimental evidence illustrating the potential of this approach. In separate work published in Cell, Krishnan and colleagues developed a generative deep-learning approach for de novo antibiotic design.[2] The researchers combined genetic algorithms and variational autoencoders with fragment-based and de novo design. Of 24 custom-synthesised compounds subsequently investigated, seven demonstrated selective antibacterial activity.[2]

Particularly interesting were two lead compounds demonstrating distinct modes of action against Neisseria gonorrhoeae and Staphylococcus aureus. These compounds also showed efficacy against multidrug-resistant strains and in mouse models.[2] This provides an important demonstration of the progression required in AI-assisted antibiotic discovery, moving from computational output to synthesis and then to experimental biological evaluation.

Yet computational sophistication should not be confused with successful translation. Krishnan and colleagues emphasise emerging data and reporting standards, translational pathways and regulatory hurdles as important elements in moving AI-discovered antimicrobials towards patients.[1]

Data quality and standardisation are consequently crucial. AI systems depend upon the datasets from which patterns are learned and predictions generated. For the field to mature, computational performance needs to be evaluated alongside robust experimental evidence and transparent reporting. The key question is ultimately not whether an algorithm can propose an interesting molecule, but whether the resulting compound possesses suitable antimicrobial and safety characteristics when subjected to experimental testing.

Economics presents an additional problem that AI cannot itself solve. The authors highlight the unfavourable economics surrounding antibiotics, where revenues tied to prescription volume make antibiotic development relatively unattractive to major pharmaceutical companies. As a consequence, academic laboratories and smaller biotechnology companies can carry considerable early-stage discovery risk while facing difficulties finding commercial partners prepared to advance candidates through clinical development.[1]

The significance of AI therefore lies not simply in making conventional antibiotic discovery faster. It creates opportunities to explore previously inaccessible chemical and sequence spaces and potentially to identify compounds with different structures and mechanisms. However, the real measure of success will be translation.

This makes the authors' concept of moving“from code to cure” especially pertinent. AI may generate increasingly sophisticated drug candidates, but computational discovery must be connected to experimental validation, appropriate data and reporting standards, viable development pathways and a workable regulatory environment.[1] The future of AI-enabled antibiotic discovery will therefore depend as much upon creating this translational ecosystem as upon developing increasingly powerful algorithms.

References

1. Krishnan A, Cao H, Collins JJ, de la Fuente-Nunez C. From code to cure: Harnessing AI for antibiotic discovery and design. Science Translational Medicine. 2026;18(871):eaee4629. doi:10.1126/scitranslmed.aee4629. PubMed record. [science.org], [pubmed.ncb...lm.nih.gov]

2. Krishnan A, Anahtar MN, Valeri JA, et al. A generative deep learning approach to de novo antibiotic design. Cell. 2025;188. doi:10.1016/j.cell.2025.07.033. [anahtarlab...arvard.edu]

Posted by Dr. Tim Sandle, Pharmaceutical Microbiology Resources (http://www.pharmamicroresources.com/)