Therapeutic peptide discovery remains limited by reliance on wild-type receptor models, fragmented evaluation pipelines, and an inability to scale across the genetic diversity of human populations. Here we introduce QuantaPep v4.0, an autonomous hybrid computational platform that unifies genotype-aware sequence design, multi-engine structural validation, and manufacturing-aware optimization within a continuous, closed-loop discovery framework. QuantaPep v4.0 integrates three core innovations. First, a distributed asynchronous discovery architecture enables high-throughput generation and evaluation of candidate sequences at scales exceeding two million candidates per campaign. Second, a machine learning-based neural scoring proxy trained on structurally validated data prioritizes candidates prior to docking, reducing computational burden while preserving predictive fidelity. Third, a self-improving optimization loop continuously refines sequence generation using aggregated validation signals, coupling statistical learning with physics-based evaluation. Applied to multi-target discovery campaigns, the platform identifies candidate peptides that satisfy simultaneous constraints on predicted binding affinity, structural stability, and manufacturability. In a representative genotype-sensitive receptor system, incorporation of variant-aware design improves predicted population coverage relative to wild-type-optimized approaches. Autonomous scaffold compression further demonstrates the platform's capacity for de novo structural optimization, producing a 12-residue triple-agonist candidate with predicted binding metrics exceeding a 30-residue reference compound by 58% in simulated affinity Z-score. All results reported herein are derived from computational evaluation pipelines and require experimental validation. Full sequences are withheld pending patent prosecution. These findings demonstrate the feasibility of autonomous, genotype-aware peptide discovery at industrial scale and establish a foundation for scalable exploration of therapeutic design spaces that have previously been computationally inaccessible.
Adam Werderits (Sun,) studied this question.