Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.21775623Overview This dataset comprises a curated collection of antibody sequences and associated functional metadata extracted from the global patent literature. By converting unstructured legal disclosures into a structured bioinformatic format, this resource enables the large-scale analysis of therapeutic antibody space, epitope landscape mapping, and the benchmarking of protein-protein interaction (PPI) prediction models. The dataset includes variable region sequences (VH and VL), target protein specifications, experimental binding affinities, and epitope residue definitions where available. Version 1.1 corrects errors found in version 1.0.1 and adds folding data for the benchmark from the paper. Dataset Structure and Documentation poisk_dataset.csv is provided in a flattened CSV format: 93,193 rows and 47 columns, each row carrying a distinct unique_record_id. For a full breakdown of the schema, refer to poisk_dataset_column_description.csv, which documents all 47 columns. Key Data Categories: Identifiers: Patent IDs, internal antibody nomenclature, and unique record identifiers. Sequences: Amino acid sequences for Heavy (VH) and Light (VL) variable regions, plus the target sequence, including associated SEQ ID NOs from the original patent filings. Target sequences carry their provenance: target.sequence is the sequence the release holds, target.sequence_from_patent what the patent's GenBank deposit holds, and target.sequence_from_uniprot what the named UniProt entry holds. Target Metadata: Target protein names, gene symbols, UniProt IDs, and source organisms. Biophysical Properties: Experimental binding affinity, assay methods (e.g., SPR, BLI), and specific binding conditions. Epitope Information: Descriptions of binding sites, including linear ranges, structural contact residues, and competition groups. Methodology: LLM-Based Extraction Extraction from high-volume, unstructured patent text was performed using an automated pipeline powered by Large Language Models. This approach captures nuanced data points — assay conditions, specific epitope residues — that are typically inaccessible to rule-based parsers. Both prompts are included verbatim, with the response schema exactly as transmitted to the model: extractor.md (one request per patent, producing the structured records) and auditor.md (one request per record, re-locating its support in the patent). Quality Control and Algorithmic Audit Users should be aware that LLM-based extraction is subject to generative inconsistencies, commonly referred to as hallucinations. Discrepancies may include the misassociation of affinity values with specific antibody variants or errors in residue indexing. To ensure data integrity we implemented a multi-stage audit layer. Consult these columns if a data point appears inconsistent: audit.verdict — one of three tiers. CONFIRMED: every field was explicitly supported by the patent text. PLAUSIBLE: nothing in the patent contradicted the record, but support was partial, indirect or inferred. REVIEW: the record appeared incorrect, or its correctness could not be established and a human should examine it. REVIEW is not an assertion that a row is false. audit.flags — field names with potential mismatches. audit.reasoning — a natural-language explanation of how the extracted value was reached. Relationship to the folding predictions poisk_all_models_folding_metrics.zip holds folding metrics, including ipTM and ipSAE, from the six models used for benchmarking: one CSV per model, 22,494 rows each, all sharing the same 28 columns (the column order differs between files, so align on names). These predictions were run against dataset version 1.0.1, not against this release. Each ipSAE-style score ships beside a column naming the chain pair it was computed on (ipsae_chain_pair, cdr_ipsae_chain_pair, and the *_ab_as_single_chain_label pair). Read that label before ranking on a score: a value whose pair is not antibody-against-antigen is not a binding metric. Usage and Limitations This dataset is intended for research and benchmarking purposes. Given the legal and technical complexity of patent filings, users are encouraged to verify critical sequences or affinity data against the original patent documents (accessible via the patent_id column) before initiating downstream experimental validation. For further details on the extraction pipeline, performance benchmarks and validation metrics, refer to the main manuscript: [Link to Manuscript/DOI].
Keyword(s):
Antibody Design; Binding Affinity; Computational Biology; Dataset; Therapeutic Antibodies.
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
No Associated Publication Data Available
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)