Why Keyword Search Fails for Biologic Patents — and How AI Solves It
You run a keyword search for “anti-TNF antibody” in a patent database. The top result is a relevant monoclonal antibody—but the patent never uses that exact phrase. It calls the drug “D2E7” in the title, “adalimumab” in claim 1, and “Humira” in the examples. Your search missed it entirely. This happens constantly in biologic patent search, where the same molecule appears under dozens of names across jurisdictions, development stages, and assignees.Sequence-based patent search matches the amino acid or nucleotide sequence of a biologic drug directly against sequences disclosed in patent documents, independent of nomenclature. Instead of searching for “anti-PD-1 antibody” and hoping every relevant patent uses that term, you submit the actual variable heavy chain (VH) sequence and retrieve every patent containing a similar sequence—regardless of whether the patent calls it pembrolizumab, MK-3475, lambrolizumab, or simply “antibody 23B.”This approach uses BLAST-style alignment algorithms from NCBI that calculate percent identity, query coverage, and e-value scores. Tools like PatSnap Biology Modality MCP—built by PatSnap, indexing 208M+ patents across 174 jurisdictions—implement this inside AI environments like Claude, enabling researchers to ask “find patents similar to this VH sequence” in plain language and receive alignment results grounded in live patent data.The fundamental difference: keyword search depends on authors choosing the same terminology you choose. Sequence search depends on molecular structure, which doesn’t change across languages, trade names, or filing strategies.
How Sequence-Based Patent Search Works
Sequence-based search operates through biological alignment: you provide an amino acid sequence (typically a VH or VL domain, full-length antibody, or complementarity-determining region CDR-H3) in standard FASTA format. The search engine compares your query against every sequence disclosed in patent documents worldwide using local alignment algorithms, returning percent identity (how many amino acids match), query coverage (what portion of your sequence aligned), and e-value (statistical significance).Query submission. Paste your sequence in standard format or raw text.Database alignment. The search engine runs asynchronously (30–90 seconds) comparing one sequence against millions in the patent corpus.Similarity scoring. Results show three metrics: a result with 95% identity across 100% query coverage in a VH domain indicates near-identical structure.Patent context retrieval. Each matching sequence links to its source patent, including assignee, legal status according to WIPO standards, claims, and filing date.Example output (anti-DLL4 antibody VH search): Sequence 1492390 matched at 92% identity across 121 amino acids (full VH coverage). The source patent lists a completely different antigen target in its title, but the framework scaffold is identical to the query—evidence of structure reuse across different antibody programs.The same search for “DLL4” as a keyword would have missed this patent entirely, because the text never links that antigen name to this particular sequence identifier.
Why Antibody Scientists and IP Teams Need Sequence Search
For researchers, sequence search solves the prior art gap that keyword methods create. Biologics exist in physical structure before they receive names. An antibody developed in 2018 under an internal code might appear in a 2015 patent under a different code, a 2020 patent under a trade name, and a Chinese filing under a transliterated variant. If you search only by name, you find only the subset of patents that happened to use your chosen term. Sequence search finds all structural matches, then you filter by legal status, assignee, or claim scope.For IP managers, this directly impacts freedom to operate (FTO) assessments. Missing a relevant patent because of naming variance is a failure mode that sequence search eliminates. When evaluating whether a candidate antibody infringes existing patents, you need to know every disclosed sequence with high similarity—not just the ones with matching keywords in the abstract.The MCP connector supports this workflow by linking sequence alignment to patent legal status and assignee data in one query. You can ask “find antibodies >90% similar to this VH and show their current legal status in the US and EU,” then filter to active patents only—no switching between BLAST results and separate patent databases. It handles antibody-antigen pair searches (all patented antibodies targeting a specific protein according to UniProt annotations), modification searches (PEGylation, glycosylation patterns), and batch sequence fetches from individual patents, all accessible through natural language prompts.The workflow shift: run a sequence-based FTO check in a single session instead of cross-referencing BLAST results against patent records manually.What Sequence Search Does Not Do
Sequence search does not interpret claim scope or provide legal conclusions. A 98% identical sequence in an expired patent is different from one in an active patent with broad claim language—the tool shows you the match and the status, but legal teams determine infringement risk.It also does not search based on functional descriptions. A patent claiming “antibodies that neutralize TNF-alpha with KD < 1 nM” without disclosing sequences cannot be found by sequence search, only by keyword or classification code searches. Sequence and keyword methods are complementary, not replacements.Try It Yourself
Start with the lowest-commitment path:1. Browser option: Go to PatSnap Eureka and paste a sequence into the chat interface. Ask “find similar sequences in patents”—results appear in seconds, no account setup required for initial exploration.2. AI workflow integration: Get a free API key at open.patsnap.com (10,000 credits, no credit card), then add the connector from the MCP marketplace to Claude or any Model Context Protocol-compatible environment. Let your AI assistant handle configuration details. Run a test sequence search to verify the connection.Both paths query the same patent and biosequence database. Choose based on whether you work primarily in a browser or inside an AI coding environment.Frequently Asked Questions
Why doesn’t keyword search work for biologic patents?
The same biologic molecule appears under multiple names across patent documents: research codes (D2E7), international nonproprietary names (adalimumab), trade names (Humira), and generic descriptors (anti-TNF monoclonal antibody). A keyword search retrieves only patents using your exact term. Sequence search retrieves every patent disclosing that molecular structure, regardless of nomenclature. This is why sequence-based methods typically find substantially more relevant prior art than keyword-only searches in antibody FTO assessments. Learn more about biologics innovation workflows.What is percent identity in sequence search results?
Percent identity measures how many amino acids match between your query sequence and a patent sequence across the aligned region. 100% means perfect match; 92% means 92 of every 100 positions are identical. For antibody VH or VL domains, >90% identity typically indicates a related scaffold or derivative. For CDR-H3 regions (the most variable part of an antibody per European Bioinformatics Institute standards), even 80% identity can signal structural similarity worth investigating, because this region dominates antigen binding specificity.Can I use sequence search for non-antibody biologics?
Yes. Sequence-based patent search works for any biologic with a defined amino acid or nucleotide sequence: therapeutic peptides, enzymes, Fc-fusion proteins, oligonucleotides, CAR-T constructs, and antibody fragments like single-chain variable fragments (scFv). The same alignment principles apply—you’re matching molecular structure against disclosed sequences in patent documents. Peptide searches typically use shorter query lengths (10–30 amino acids), while full antibody searches might align 400+ amino acids across heavy and light chains.Do I need a paid subscription to try sequence-based patent search?
No. You can start immediately with PatSnap Eureka in your browser at no cost, or create a free account at open.patsnap.com for 10,000 credits (no credit card required). Both access the same global patent and biosequence database. Pay-as-you-go pricing applies only when you exceed free credits. For MCP integration, you’ll need the API key from account signup, then add the connector from the marketplace—setup takes minutes.Note: Information based on publicly available sources as of 2026. Product features may change. Contact PatSnap for current specifications.
