PeptAI says its agent independently redirected its search for a peptide targeting VEGFR2 after a laboratory test
PeptAI says its agent independently redirected its search for a peptide targeting VEGFR2 after a laboratory test
According to the company, the program generated thousands of candidates, selected several, paid for binding measurements on its own, and received the results. PeptAI says these data prompted the system to change its peptide generation method and exclude problematic regions of the target in the next round.
In April, PeptAI presented nine computational checks that the program applies to peptides before sending them to an external laboratory. In May, its agent proposed 306 MOTS-C peptide candidates and an experimental plan, but none of the candidates had yet been synthesized.
On 17 July, PeptAI wrote in a new report:
“The agent designs a candidate, pays a laboratory to test it, receives the binding result, and then begins the next round on its own.”
The company says the agent generated thousands of peptides targeting VEGFR2, selected several, and paid for laboratory testing. VEGFR2 is a receptor on vascular cells that transmits signals promoting the growth of new blood vessels, including vessels that supply tumors.
PeptAI reports that the positive control and a deliberately modified variant bound to the target, while the negative control produced no signal. The agent used this result as feedback. The next round will use BindCraft, a program for designing proteins that bind to a specified target, and will avoid known problematic regions on the surface of VEGFR2.
The company remains the sole source of these data. The post provides no sequences, protocol, numerical measurements, or independent confirmation. A binding assay answers a narrow question: whether a peptide binds to the target in that specific assay. A drug development program would then require evidence of activity in cells and organisms, as well as data on selectivity, safety, and reproducibility.
A model can rank thousands of sequences, but a physical experiment reveals where its prediction encountered the actual molecule. For the agent, each “sequence and result” pair becomes a record of which ideas survived testing. If the result genuinely changes the next search, every success and failure narrows the space explored in the following experiment.