De Novo Protein Design Patents

De novo (AI) protein design patents cover AI-design-model innovations; designed-protein (composition) innovations; design-platform/workflow innovations; and §101-eligibility and validation innovations — with IP held by generative-biology companies and the foundational academic labs (in a field that uses AI to create entirely new proteins from scratch). WHY DE NOVO PROTEIN DESIGN: AlphaFold solved PREDICTING the structure of EXISTING proteins; DE NOVO design goes further — using generative AI to CREATE entirely NEW proteins that don't exist in nature, with custom shapes and functions (novel BINDERS to disease targets, new ENZYMES, vaccines, biomaterials) — vastly expanding what can be drugged or engineered beyond nature's repertoire; a transformative tool for drug discovery and biotechnology. CRITICALLY, like AlphaFold, the AI MODELS are largely OPEN-SOURCE/published (David Baker's lab releases RFdiffusion and others freely), so the durable IP is mostly NOT in the models but in the specific DESIGNED PROTEINS and their applications. MAJOR HOLDERS: University of Washington / DAVID BAKER LAB (RFdiffusion, RoseTTAFold — foundational, much open-sourced; Baker won a Nobel for protein design), GENERATE BIOMEDICINES (Chroma), XAIRA, EVOLUTIONARYSCALE (ESM), ARC/PROFLUENT, CRADLE, plus pharma. AI design models, designed-protein compositions, design platforms/workflows, §101 eligibility, and validation are the core de-novo-design patent domains — and designed proteins, platforms, and applications are the open whitespace (models are largely open).

Designed-protein-composition innovations; §101-eligibility advantages; novel-binder/enzyme innovations; and therapeutic-application innovations represent core de-novo-design patent domains — and the specific man-made protein (which is more clearly patent-eligible than a natural one) is the foundational, high-value asset. DESIGNED-PROTEIN COMPOSITION PATENTS: the specific NOVEL protein created by the design process — its SEQUENCE and structure — for a particular function (a binder, enzyme, scaffold, vaccine antigen); the designed protein is core COMPOSITION-OF-MATTER IP (the protein is the product/drug — this is where the durable value is, NOT in the largely-open models). §101-ELIGIBILITY ADVANTAGE PATENTS: a key point — a DE-NOVO-DESIGNED protein is NOT a 'product of nature' (it's man-made, doesn't exist in nature), so it AVOIDS the Myriad natural-product §101 problem that limits patenting isolated natural proteins — making de novo designs MORE clearly patent-eligible than natural biomolecules (a real strategic advantage); claiming strategy leverages this man-made distinction. NOVEL-BINDER / ENZYME PATENTS: specific applications — designed BINDERS (novel proteins that bind a disease target with custom affinity/specificity, as therapeutics or research tools — a huge use), novel ENZYMES (designed catalysts for industrial/therapeutic reactions), and novel scaffolds; binder/enzyme compositions are high-value, defensible IP. THERAPEUTIC-APPLICATION PATENTS: designed proteins as DRUGS (therapeutic binders, novel-format biologics, vaccine immunogens) and their uses; therapeutic-application compositions/methods are high-value. Designed-protein compositions, the §101 advantage, novel binders/enzymes, and therapeutic applications are the highest-value core IP because the specific man-made protein — patentable in a way natural proteins aren't — is exactly where de novo design's durable value lives.

AI-design-model innovations; design-platform/workflow innovations; validation/developability innovations; and data and §101-software considerations represent additional de-novo-design patent domains — and the models, the design pipeline, and proving designs work are where patentability is mixed (models are often open) and value is mostly in execution. AI-DESIGN-MODEL PATENTS: the generative MODELS that create proteins — DIFFUSION models (RFdiffusion, Generate's Chroma), protein LANGUAGE MODELS (ESM, ProtGPT), and structure-based generators; BUT many foundational models are OPEN-SOURCE/published (Baker lab), so model patentability is LIMITED and §101 (abstract-idea/math) applies to algorithms — novel, non-open model architectures or training methods may be patentable, but the model is often NOT where defensible IP lies (like AlphaFold — open models, value in outputs); claim concrete technical methods if pursuing model IP. DESIGN-PLATFORM / WORKFLOW PATENTS: the end-to-end DESIGN-build-test PLATFORM — combining generative design with screening, filtering, optimization, and experimental feedback loops (active learning) — and proprietary methods making designs succeed more often; platform/workflow methods are valuable (the integrated pipeline and know-how are a moat even when models are open). VALIDATION / DEVELOPABILITY PATENTS: methods ensuring designed proteins actually FOLD, FUNCTION, are stable, and are MANUFACTURABLE (many AI designs fail experimentally — the design-success rate and developability are key) — and computational filters predicting developability; validation/developability methods are valuable. DATA / §101 PATENTS: proprietary experimental DATA (to train/fine-tune better-than-open models — often the real moat) and §101-aware claiming. AI design models (limited by open-source), design platforms/workflows, validation, and data are the highest-value execution IP because — with models largely open — the design pipeline, success rate, validation, and data are exactly what differentiate (alongside the designed proteins themselves).

De novo protein design startup IP strategy must navigate the OPEN-SOURCE model reality (foundational models — RFdiffusion etc. — are largely free/published, so models are mostly NOT defensible IP, like AlphaFold), the Baker-lab/academic foundational IP, generative-biology company portfolios (Generate/Xaira/EvolutionaryScale), the composition-of-matter primacy (the designed PROTEIN is the durable, patentable asset), the §101 ADVANTAGE (de novo proteins are man-made — more clearly patent-eligible than natural proteins — a real plus to leverage), the §101 model/software limits (algorithms face abstract-idea issues), the validation/developability challenge (many designs fail — execution matters), the data moat (proprietary experimental data trains better models), the heavy clinical/FDA path for therapeutics, and a landscape where designed proteins, platforms, validation, and data are the durable assets; understand that models are largely open, so the durable IP is in the DESIGNED-PROTEIN compositions (binders/enzymes/therapeutics — leveraging the §101 man-made advantage), the design platform/workflow, validation/developability, and proprietary data — with the proteins themselves and execution/data often the real moat, and that the value of designed proteins, design success rate, clinical efficacy, and data matter as much as model patents; identify whitespace in designed proteins, platforms, and applications. DE-NOVO-DESIGN STARTUP IP STRATEGY: MODELS ARE LARGELY OPEN — DESIGNED-PROTEIN COMPOSITIONS, DESIGN PLATFORMS/WORKFLOWS, VALIDATION, AND DATA ARE THE IP: patent the specific designed PROTEINS (binders/enzymes/therapeutics), design platforms/workflows, validation/developability methods, and protect proprietary data — don't rely on the (often open) models for defensibility; THE DESIGNED PROTEIN IS THE DURABLE ASSET (COMPOSITION-OF-MATTER): the specific novel protein/sequence for a function is the core, patentable IP — this is where value lives (like AlphaFold: open models, value in outputs); LEVERAGE THE §101 MAN-MADE ADVANTAGE: de-novo-designed proteins are NOT natural products, so they AVOID the Myriad natural-product §101 problem and are MORE clearly patent-eligible than natural proteins — a real strategic advantage in claiming; MODELS FACE OPEN-SOURCE + §101 LIMITS: foundational models are free/published and algorithms face abstract-idea issues — model IP is mostly NOT defensible (novel non-open architectures aside); DESIGN PLATFORM/WORKFLOW + SUCCESS RATE IS A MOAT: the integrated design-build-test pipeline (and getting designs to actually work) is a real, defensible advantage even with open models; VALIDATION/DEVELOPABILITY IS THE EXECUTION CHALLENGE: many AI designs fail experimentally — methods raising design success/developability are valuable; PROPRIETARY DATA IS OFTEN THE REAL MOAT: experimental data to train/fine-tune beyond open models drives better designs — a key (often trade-secret) advantage; APPLICATIONS (BINDERS/ENZYMES/THERAPEUTICS) DRIVE VALUE: designed binders to disease targets and novel enzymes are high-value composition IP; PROTEIN-VALUE/SUCCESS-RATE/CLINICAL/DATA MATTER AS MUCH AS PATENTS: the value of the designed proteins, design success rate, clinical efficacy, and data drive value (models alone don't); WHEN TO PATENT: NOVEL DESIGNED PROTEIN/PLATFORM/VALIDATION WITH MEASURED DATA: file once a designed protein shows measured results (binding/affinity/specificity or catalytic activity + folding/stability + developability + function/efficacy) — the designed protein's measured function, the §101 man-made advantage, and design success/developability are the critical de-novo-design IP metrics; KEY FTO CHECKLIST: Baker lab/UW (RFdiffusion/RoseTTAFold — much open); Generate (Chroma)/Xaira/EvolutionaryScale (ESM)/Profluent/Cradle; designed-protein composition-of-matter (sequence/structure/function); §101 man-made advantage (vs Myriad natural product); novel binders/enzymes/scaffolds/vaccines; AI design models (diffusion/language, open-source + §101 limits); design platform/workflow/active-learning; validation/developability/fold-function/manufacturability; proprietary experimental data (trade-secret); therapeutic applications/FDA path.