Peptide Design in 2026: AI Predictions and the Experimental Evidence Gap

## Two recent approaches to computational design Two preprints posted in June 2026 offer a useful view of how researchers are organizing peptide-design problems. One focuses on tracking the decisions made during iterative sequence optimization. The other explores a computational representation for generating macrocyclic peptides. Both are early research reports, and the findings discussed here should be read as preprint evidence rather than established experimental conclusions. ## Making sequence optimization inspectable In *Pepti-Agent*, submitted June 13, 2026, Houxu Chen and colleagues describe a framework that separates candidate generation, property prediction, and sequence mutation into inspectable tools. A controller consults predictor outputs while refining candidates, and the system records decisions and accepted sequence changes along the way. The stated purpose includes comparing design strategies and prioritizing candidates for experimental validation. [Chen and colleagues: Pepti-Agent, preprint](https://arxiv.org/abs/2606.15422) For journal readers, an interesting feature is the emphasis on the decision record. A candidate sequence is easier to evaluate when the reasoning path, scoring steps, and modifications can be examined. That is an editorial observation about research transparency; it is not evidence that the resulting sequence has the predicted properties in a laboratory assay. ## Generating macrocyclic structures *PepALD*, submitted June 12, 2026, describes a different design approach. The authors use representations of chemical building blocks and a generative model that accounts for ring closure while constructing macrocyclic peptides. Its abstract reports computational experiments evaluating generation and optimization against comparison methods. The results summarized there are explicitly in silico. [Zhang and colleagues: PepALD, preprint](https://arxiv.org/abs/2606.14510) That evidence category matters. A computer-generated structure, an experimentally characterized molecule, and a reproducible assay result represent different stages of investigation. Reporting should identify the stage actually reached. ## What the next evidence should address Useful follow-up questions include whether proposed candidates can be synthesized and characterized, whether measured properties agree with predictions, and whether independent experiments reproduce the findings. Transparent selection criteria also matter: readers need to know whether a displayed result represents a broad evaluation or a small set of selected examples. These questions provide a framework for following future publications without turning preliminary results into promises. The developments are relevant to research methods and evidence quality; neither preprint establishes the identity, purity, or performance of materials in The Peptide Gallery's catalog. The most informative next update will connect a clearly described design process with equally clear experimental observations.