A hierarchical reaction logic approach enabled a computer algorithm to design complete synthesis routes for complex peptides like vancomycin and semaglutide within minutes, matching strategies used by human experts.
Minutes to design routesThe algorithm planned complete synthesis routes for peptides as complex as vancomycin and as large as semaglutide in just minutes, without any training on published procedures.
What the researchers found
By constraining computer-assisted synthesis planning with hierarchical reaction logic — dictating which subsets of reactions to apply at different planning stages — the algorithm successfully designed complete synthesis routes for complex peptide targets within minutes.
The approach was validated on clinically relevant targets including vancomycin (a complex glycopeptide antibiotic) and semaglutide (a large GLP-1 receptor agonist used for diabetes and obesity). Despite not being trained on any literature precedents, the computationally designed routes mimicked strategies used by human expert chemists.
The system incorporates protecting-group strategies and realistic pathway pricing, and supports both solid-phase and solution-phase synthesis modes, including C-to-N and N-to-C peptide extension strategies.
Why it matters
Peptide drug manufacturing is one of the biggest bottlenecks in bringing peptide therapeutics to market. Designing synthesis routes for complex peptides currently requires deep expertise and significant time. An algorithm that can produce expert-quality synthesis plans in minutes could accelerate drug development, reduce manufacturing costs, and make complex peptide drugs more accessible.
How the study worked
The researchers developed a hierarchical planning framework that layers reaction logic on top of existing retrosynthetic search algorithms. Rather than considering all possible reactions at every step, the system applies specific subsets of reaction transforms at different stages of route planning. The algorithm was tested on complex peptide targets without any training on published synthesis routes, and the resulting routes were compared to human expert strategies.
What this study cannot tell us
The abstract does not report whether the computationally designed routes were actually executed in the laboratory to validate their practical feasibility. Computational route planning may not account for all real-world challenges such as side reactions, purification difficulties, and scalability issues. The comparison to human expert strategies is qualitative. The algorithm's performance on targets beyond vancomycin and semaglutide is not detailed in the abstract.
How to read the evidence
This is a computational methodology study published in JACS (Journal of the American Chemical Society), a top-tier chemistry journal. The results demonstrate algorithmic capability but experimental validation of the designed routes is not described in the abstract.
When this study was published
Published in 2025, this is cutting-edge research applying AI to peptide synthesis at a time when demand for complex peptide manufacturing is surging.
The bigger picture
This work sits at the intersection of artificial intelligence and pharmaceutical manufacturing. As peptide therapeutics become increasingly important — with drugs like semaglutide generating billions in revenue — efficient synthesis planning becomes commercially critical. The hierarchical logic approach also represents a broader insight for AI: sometimes constraining an algorithm with domain knowledge produces better results than giving it unlimited data, a principle applicable beyond chemistry.
Questions still open
- Have any of the computationally designed synthesis routes been validated experimentally in the lab?
- Could this hierarchical approach be extended to other complex biomolecule classes like proteins or nucleic acids?
- How does the algorithm handle synthesis route optimization for cost and yield at industrial manufacturing scale?
Common questions
Why is it so hard to make complex peptides like semaglutide?
How does this algorithm differ from other AI chemistry tools?
Read the original research
Hierarchical Reaction Logic Enables Computational Design of Complex Peptide Syntheses.
Journal of the American Chemical Society, 147(9), 7644-7662
Citation
Molga, Karol; Beker, Wiktor; Roszak, Rafał; Czerwiński, Andrzej; Grzybowski, Bartosz A. (2025). Hierarchical Reaction Logic Enables Computational Design of Complex Peptide Syntheses.. Journal of the American Chemical Society, 147(9), 7644-7662. https://doi.org/10.1021/jacs.4c17057