peptides that don't work

simulating peptide changes, and what compounds will not work with the bounding pockets. The bench loads a protein structure, finds the pockets on its surface, and proposes peptides that should fit: designing new ones, and searching 17,000 real catalogued ones, scored on the same objective so you can see which wins. You drive it in plain English — one question, and GPT-6 Astra calls the tools, runs the physics, and pulls trial results for drugs already approved against that target.