Generative Hit-DesignIndoAutoImmune 2.0 extensionTwo tracks · Five gates · One CRO handoff package
The target space competitors aren't looking in.
Autoimmune targets with a distinct India-specific genetic signal that also lack an approved drug — surfaced from the same graph that powers the rest of PetriDish, then handed to a dual-track generative pipeline. Small-molecule track (DrugDiff-class) for TYK2-family enzymes; binder track (RFdiffusion3 + ProteinMPNN) for HLA-peptide grooves. Every candidate passes five hard go/no-go gates before it enters the CRO handoff package.
📦
CANDIDATE PACKAGE FOR CRO / PHARMA R&D PARTNER · NOT A DRUG
Per PRD §1 + §9: this module explicitly stops at "ranked shortlist a partner finds credible enough to fund." It does not replace, deliver, or predict wet-lab validation, preclinical toxicology, IND-enabling studies, or clinical trials. Field-wide base rate: ~90% of drug candidates that reach human trials still fail, AI-designed or not (PRD §8). Overclaiming here damages credibility with exactly the pharma partners you need.
Two design tracks (PRD §2.1)
SMALL-MOLECULE TRACK
TYK2-class enzyme targets
Structure → pocket detection → DrugDiff latent diffusion → Boltz-2 co-folding → ADMET → ranked shortlist
Precedent: Nimbus/Schrödinger zasocitinib (TAK-279 Phase III); Atomwise independent TYK2 candidate
PROTEIN BINDER TRACK
HLA-peptide-modulating targets
HLA groove → RFdiffusion3 atom-level backbone → ProteinMPNN inverse fold → Boltz-2 re-validation → ranked shortlist
Precedent: Directly extends IndoAutoImmune 2.0 HLA imputation + structural work
PRD §2 target discovery · India-signal + white-space filter
Surfaces autoimmune targets where (a) India-specific genetic signal is elevated relative to global, (b) pathway is autoimmune-relevant, (c) no approved drug exists yet. This is the search space competitors aren't looking in.