AI Protein Fitness Landscape Explorer

AI-driven protein design rarely optimizes one number — a sequence that binds tighter is often less stable or more prone to aggregate. This simulator renders that trade-off as a real 3D terrain: every candidate sequence is a point scored by simplified stability (ΔΔG), binding-affinity (pK_D) and aggregation-propensity models, combined into one fitness height. Each "Run cycle" click plays out one round of a design-build-test loop — the AI proposes a new pool of candidates around the current best, scores them, filters out the ones that fail a developability cutoff, and promotes the new peak — while live readouts track how many candidates pass, and the predicted K_D and ΔΔG of the current leader.