AI designs components of "microorganisms that produce medicine," increasing production by up to approximately three times.

AI Designed Components for a "Drug-Making Factory"

When searching for new drugs, finding a promising substance and producing it in sufficient quantities are separate challenges.

Natural products produced by microorganisms can lead to pharmaceutical development. However, modifying substances found in nature to suit human treatment and securing the necessary quantities for research is not easy due to the complexity of the cellular mechanisms that produce these substances.

Research is underway to redesign parts of this production apparatus using artificial intelligence.

A research team at the Helmholtz Institute for Pharmaceutical Research Saarland (HIPS) in Germany used AI to design protein components that work inside large enzymes assembling natural products. They tested these in living cells and demonstrated that some functioned. In specific enzyme systems, production increased by up to about three times compared to using naturally derived components.

This is not a story about AI creating a finished new drug. It is research showing the potential to improve the "manufacturing mechanisms" that microorganisms possess to search for drug candidates.

The Precise "Molecular Assembly Line" Inside Microorganisms

The research focused on a large group of enzymes called "non-ribosomal peptide synthetases," abbreviated as NRPS.

While the terminology may be complex, their function can be likened to a factory assembly line. Multiple components share roles, receive materials, process them, and pass them to the next stage. Through their collaboration, distinctive compounds are produced.

Natural products made by NRPS have been the basis for developing antibiotics, immunosuppressants, and anticancer drugs. Therefore, if this mechanism can be freely reconfigured, it might be possible to create substances with different properties from conventional ones.

However, biological assembly lines cannot be easily swapped like commercial blocks.

Changing the design of one part can disrupt its coordination with adjacent parts, halting material transfer and rendering the entire enzyme non-functional. Even if a component is well-formed on its own, it may not function within the production apparatus.

AI Designed the "Transporter" for Material Transfer

Dr. Kenan Bozhüyük's team focused on the "thiolation domain," or T domain, within NRPS.

This domain is responsible for receiving molecules during production and transporting them to the next processing site. In a factory analogy, it resembles a movable transport arm connecting processes.

The research team combined multiple pre-trained AI models and experiments to design 76 new T domains. They incorporated these components into various NRPS and tested hundreds of enzyme variations in living cells.

The key point is that they did not end with computational evaluations of AI-proposed designs. They actually created, integrated, and examined whether the products could be generated.

As a result, some AI-designed components functioned. They enabled substance production in artificially combined enzyme systems and increased production levels. At least one designed protein was found to be more soluble and stable against heat compared to its natural counterpart.

What Does "Up to About Three Times" Mean?

The attention-grabbing result of this achievement is the figure indicating production increased by up to about three times.

However, this result was not obtained for all designed components or substances. It was an improvement confirmed under specific enzyme systems and conditions, not an indication that drug efficacy tripled or drug development time was reduced by a third.

Nevertheless, the improvement in production has research significance.

Even if a new substance is created, if the quantity is too small, it becomes difficult to examine its structure and properties in detail. Securing sufficient quantities makes it easier to verify what kind of molecule it is and its effects.

Thus, technology that enhances production efficiency is not only relevant to future mass production but is also crucial for making candidate substances "thoroughly examinable" in the early stages of drug development.

The Key to Success Lies in "Compatibility with Neighbors" Rather Than Component Performance

An important aspect of understanding this research is how the designed components were adapted to their surrounding environment.

Even if the transporter is stable on its own, the assembly line won't function if it cannot properly interact with the recipient or the next stage. It is necessary to transfer the molecule being produced to the appropriate location.

The research team focused not only on treating the T domain as an independent protein but also on how it interacts with adjacent regions within the enzyme it is incorporated into.

This highlights the difficulty of using AI in biological design. There is a gap between creating "plausible components" and components that work within "complex apparatuses."

Dr. Bozhüyük explains that AI is not a replacement for experimentation but helps efficiently select promising candidates from many options. Advancing research involves the iterative process of design and experimentation.

Researchers Express Optimism on Social Media, Emphasizing Conditional Nature of Results

Social media posts related to this research were primarily from the researchers themselves or related research organizations on LinkedIn. There is no evidence that it has become widely discussed among general users, so here we introduce posts directly related to the research.

Dr. Bozhüyük introduced the results at the pre-peer-review stage, explaining the design of 76 T domains and the construction and testing of 578 recombinant NRPS in cells. He also clarified that the improvement of up to about three times in production is dependent on surrounding configurations.

He also sought opinions on the design considering the junctions and interacting partners of components. Instead of simply promoting success, he aimed to discuss the conditions under which the design works.

The related research organization PharmaScienceHub also reshared this research, expressing expectations for future developments.

These are not the impressions of general readers gathered after the news release in October 2026, but rather communications within the research community from the preprint stage. Considering this scope, the focus is more on the increased means to improve complex enzymes than the expectation of "AI alone creating drugs."

What Is Important for Drug Discovery in Japan?

Considering this achievement for Japan, the focus should be on the connection between technology for predicting candidate substances and the technology to actually produce them.

Even if a substance is attractive in calculations, if it cannot be produced in sufficient quantities, further evaluation becomes difficult. If the production apparatus can be improved as in this case, substances that were previously difficult to secure could become research targets.

Moreover, if it becomes possible to selectively produce substances with slightly different structures starting from natural products, the range of candidates that can be compared and studied will expand. However, this achievement alone does not prove that such developments will be widely realized.

When considering applications by Japanese universities and pharmaceutical companies, AI's proposal capabilities alone are insufficient. It is important to connect technologies for handling microorganisms, evaluating proteins, and analyzing the produced substances.

This research indicates that the ultimate place to verify the value of AI is also in the laboratory.

The Path to New Drugs Continues, Yet Lowering the "Barrier to Production" Holds Value

This report is foundational research for improving biological production apparatuses. It does not confirm the efficacy or safety of new therapeutic drugs in humans.

Future challenges include whether the same method applies to other enzyme systems, which combinations of components will function, and how to connect to stable production. If the produced substances become drug candidates, further detailed verification of their effects and safety will be necessary.

Even so, the significance of having more options than just using production apparatuses found in nature is substantial.

AI designs small transport components to obtain complex molecules produced by microorganisms more flexibly and in more researchable quantities.

The future of drug discovery may change not only from the discovery of prominent "new drug candidates" but also from the improvement of such components that make it possible to actually produce those candidates.

Sources

1. Phys.org:
   Published on October 21, 2026. Introduction of AI-designed T domains by HIPS, verification in cells, production improvement, and explanation by researchers.
   https://phys.org/news/2026-10-ai-functional-enzyme-components-drug.html

2. Nature Communications: Research Paper Publication Information
   Emre F. Bülbül et al. "Generative AI designs functional thiolation domains for reprogramming non-ribosomal peptide synthetases." Published on September 22, 2026. Different from the publication date of the research introduction article.Nature Communications
   https://doi.org/10.1038/s41467-026-77963-6

3. Kenan Bozhüyük's LinkedIn Post: SNS Communication by the Researcher
   Introduction of preprint stage results. Confirmation of 76 designs, 578 recombinant NRPS tests, production improvement dependent on conditions, and solicitation of opinions on design.Kenan Bozhüyük
   https://www.linkedin.com/posts/kenan-bozh%C3%BCy%C3%BCk-16a29924b_proteindesign-syntheticbiology-nrps-activity-7435231632892207104-qbGo

4. PharmaScienceHub's LinkedIn Post: Reaction by Related Research Organization
   Post showing resharing of the research and expectations for future developments. Introduced as a reaction from a research-related organization, not general public opinion.PharmaScienceHub (PSH)
   https://www.linkedin.com/posts/pharmascihub_were-excited-to-reshare-a-new-preprint-by-activity-7435264559105417216-677m