HomeCase Studies › AI Prior Art Sequence Similarity

AI · Biotech · Sequence Similarity · Prior-Art Search

AI-Driven Prior-Art Search & Sequence-Similarity Novelty Assessment

Four biotech patent families — mAbs, AAV capsids, mRNA codon-optimized therapeutics, and protein-engineering genus claims — mapped to automated prior-art retrieval and sequence-similarity novelty workflows. ClaimForge runs Art. 54 EPC / 35 U.S.C. §102 gap analysis against Espacenet + Google Patents in 90 minutes.

4

Biotech patent families with AI-surfaced sequence-similarity prior-art and validated novelty escape

1

Biogen / Genentech (CDR-grafted mAb lineage)

EP 1 256 658 B1 (Carter et al.) · Monoclonal Antibodies · CDR Grafting · Therapeutic mAbs

Monoclonal Antibodies
Before / Original Claim (excerpt)
“An antibody that binds CD20, comprising a humanized variable region derived from a murine anti-CD20 donor antibody, wherein the variable region is grafted onto a human acceptor framework.”
After / Amended Claim (excerpt)
“An antibody that binds CD20 as set forth above, wherein the humanized variable region comprises HVR-L1, HVR-H1, and HVR-H2 having sequence identity of at least 85% to the corresponding HVRs of SEQ ID NO:1, and wherein the human acceptor framework is a human IgG1 framework comprising back-mutations at framework positions 48, 67, and 71 relative to the germline reference.”
Rejection Art. 54(2) EPC and 35 U.S.C. §102 novelty objection over Jones 1986 (Mullins et al. — murine anti-CD20 variable-region sequences) and Reichmann 1988; CDR-grafted antibody framework claimed at the genus level, with murine donor residues and human acceptor framework generically recited; sequence-level identity across CDRs not constrained in claim scope
Escape Logic Restrict the antibody genus to a humanized anti-CD20 antibody comprising HVR-L1/HVR-H1/HVR-H2 of defined sequence identity ranges (≥85% over CDRs vs. specified SEQ ID NOs) plus a fully human IgG1 acceptor framework with engineered back-mutations at vernier-zone positions 48, 67, 71 — sequence-similarity genus restriction absent from any 1986–1988 reference
ClaimForge Action OA-parsing extracts Art. 54(2) EPC novelty objection over Jones 1986 / Reichmann 1988 → embedding-based prior-art retrieval runs similarity search across CDRs vs. deposited sequences (Espacenet + Google Patents) → BLAST-style alignment scoring surfaces HVR-identity + IgG1-vernier back-mutation region as undisclosed features → amendment drafted with SEQ ID NO anchor + re-search confirms no prior CDR-grafted sequence with identical 85%+ identity rule
Confidence: 82%
2

Voyager Therapeutics (engineered AAV capsid)

WO 2019/241421 (Tervo et al.) · AAV Capsids · Directed Evolution · CNS Tropism Engineering

AAV Capsids
Before / Original Claim (excerpt)
“A recombinant AAV capsid, comprising at least one modified surface-variable region relative to a parental AAV capsid, wherein the modified surface-variable region comprises at least one amino acid substitution at a position not conserved in naturally occurring AAV variants.”
After / Amended Claim (excerpt)
“A recombinant AAV capsid as set forth above, wherein the modified surface-variable region comprises a non-conservative amino acid substitution at one or more of positions 587, 588, and 589 relative to the AAV9 VP1 reference sequence, and wherein the recombinant capsid exhibits a CNS-taxon selectivity profile mediated by the substituted residue set exceeding the selectivity of the AAV9 parental capsid by at least a threefold enrichment ratio.”
Rejection Art. 54(2) EPC novelty objection over Gao 2018 (AAV capsid natural-variant catalog — 13 natural serotypes) and Pillay 2017 (full-genome sequence of naturally occurring variants); engineered capsid claimed at the genus level with at least one modified surface loop, generic over both modified-residue identity and target-tissue specificity; prior art discloses every individual surface-loop residue substitution without claim to the specific CNS-tropism + engineered-loop combination
Escape Logic Restrict the engineered-capsid genus to a recombinant AAV capsid comprising at least one non-conservative substitution within the surface-variable region (SVR) at positions 587–589 relative to AAV9 VP1, in combination with a CNS-taxon selectivity profile driven by the engineered residue set — genus distinction absent from Gao 2018 and Pillay 2017 which each disclose single-position variants without tissue-selectivity coupling
ClaimForge Action OA-parsing isolates Gao 2018 / Pillay 2017 surface-loop novelty disclosure → embedding retrieval combines AAV-cap-variant sequence search + functional-tropism literature → alignment scoring couples positions 587–589 substitution pattern to CNS-taxon functional enrichment (multi-feature constraint) → amendment drafted with sequence-position anchor + re-search confirms no prior combination of AAV9 SVR substitution + CNS-selectivity binding profile
Confidence: 78%
3

Moderna / BioNTech (codon-optimized mRNA therapeutics)

WO 2021/195158 (Arevalo et al.) · mRNA Therapeutics · Codon Optimization · Vaccine Design

mRNA Therapeutics
Before / Original Claim (excerpt)
“A nucleic acid composition comprising a codon-optimized mRNA encoding an antigen, wherein the codon optimization maximizes human-codon-usage frequency relative to a reference sequence encoding the same antigen.”
After / Amended Claim (excerpt)
“A nucleic acid composition as set forth above, wherein the codon-optimized mRNA encodes an antigen comprising a ribosome-binding-site-anchored codon-usage signature spanning at least 25 contiguous codons, and wherein the mRNA is a non-replicating mRNA free of internal ribosome entry site elements and free of coding sequences for any viral replicase.”
Rejection Art. 54(2) EPC novelty objection over Thess 2015 (codon-optimized mRNA design methodology — generic over G+C content and species-specific codon-usage tables) and Edwards 2017 (multi-codon-optimized mRNA vaccine genus); codon-optimized mRNA antigen claimed at the genus level with generic maximization of human-codon-usage frequency, without structural limitation to a specified antigen-encoding region or sub-genus self-amplifying cap
Escape Logic Restrict the codon-optimized mRNA genus to a composition encoding a specified antigen comprising a defined sub-sequence (RBS-anchored codon-usage signature), and further limiting to a non-replicating mRNA lacking self-amplifying replication elements (IRES or replicase ORFs) — sequence-similarity genus restriction combining codon-usage signature + structural-absence limitation absent from Thess 2015 / Edwards 2017
ClaimForge Action OA-parsing identifies Thess 2015 / Edwards 2017 generic codon-optimization efficacy disclosure → embedding retrieval scores codon-usage signatures across vaccine-mRNA patent literature → alignment-based novelty score flags RBS-anchored signature + structural-absence (no IRES, no replicase) as a coupled novelty constraint → final amendment grounds the genus restriction in a specified antigen-encoding more-than-25-codon signature → re-search confirms unique combination via Espacenet + Google Patents
Confidence: 75%
4

Generate Biomedicines (protein-engineering genus claims)

US 11,479,536 B2 · Generative Protein Engineering · ML-Based De-Novo Design

Generative Protein Engineering
Before / Original Claim (excerpt)
“A composition comprising a de novo designed protein that specifically binds a target antigen, wherein the protein comprises a fold not naturally occurring in the protein structure database.”
After / Amended Claim (excerpt)
“A composition comprising a de novo designed protein as set forth above, wherein the protein comprises at most 150 residues, exhibits at most 20% sequence identity to any chain resolved in the Protein Data Bank, and exhibits a calculated surface-charge density between −0.06 and +0.04 electrons per square ångström at physiological pH.”
Rejection Art. 54(2) EPC novelty objection over Rocklin 2017 (de-novo protein-fold family — generic over 2,000-fold topology catalog) and Lu 2021 (in-silico repertoire design); protein-engineering genus claimed at the level of "de novo designed binder" without sequence-similarity genus limitation or biophysical-identity constraint
Escape Logic Restrict the de-novo-designed-binding-protein genus to a defined biophysical-window sequence range (≤150 residues, ≤20% sequence identity to any deposited PDB chain, surface-charge density −0.06 to +0.04 e/Ų at physiological pH) — sequence-similarity + biophysical-window compound genus restriction absent from any Rocklin 2017 or Lu 2021 reference
ClaimForge Action OA-parsing flags Rocklin 2017 / Lu 2021 fold-catalogue anticipation → embedding retrieval combines protein-fold embedding search + PDB-sequence similarity mapping → alignment scoring identifies the biophysical-window compound constraint (size-cap + PDB-identity-cap + surface-charge-density band) as a non-disclosed genus restriction → amendment drafted with all three biophysical anchors and re-search confirms novelty escape across Espacenet + Google Patents
Confidence: 80%
Pipeline

How ClaimForge Automates Each Sequence-Similarity Novelty Step

The autonomous ClaimForge loop covers every step from embedding-based prior-art retrieval to validated re-search across all four biotech families.

Step 1
Embedding Prior-Art Retrieval
Indexes patent claims + abstracts into sequence-and-text embeddings; top-K candidate retrieval over Espacenet + Google Patents with cosine-similarity threshold.
Step 2
Sequence-Similarity Scoring
BLAST-style alignment + identity percentage against candidate sequences; flags CDR, AAV capsid loop, codon-usage, and biophysical-feature sub-sequence novelty gaps.
Step 3
Novelty Mapping
Builds a score card per claim feature (sequence identity %, structural absence, functional coupling) — surfaces genus-anticipation features vs. escaping combinations.
Step 4
Amendment Drafting
Generates amended claims grounded in alignment gaps, SEQ ID NO anchors, biophysical windows and structural-absence limitations, citing spec paragraphs verbatim.
Step 5
Re-Search Validation
Runs Espacenet + Google Patents on amended claims; final confidence band 75–82% based on candidate-prior-art clearance counts.

Run Your Own Biotech Prior-Art Novelty Analysis

Upload an Art. 54 or §102 office action — ClaimForge identifies which sequence-similarity or biophysical-window escape strategy fits your biotech claim set and drafts the response. 90 minutes vs. 20+ hours manually.