Biological Sequence Extraction & Normalization
Extract the protein and nucleic acid sequences from patents, literature, and lab records, normalize their format, decompose their modification information, and align the numbering system with body-text citations. A sequence looks like just a string, but its value depends on whether it can be re-matched to its number, modifications, and source—if it can't be, the sequence is unusable.
Where This Kind of Task Gets Hard
A sequence is not just that string of letters1. Sequences are inherently ill-suited to being "recognized"
Protein and nucleic acid sequences are long strings mixing letters and digits, with none of the redundancy that natural language has for error correction. If one residue is read as its neighbor, the biological meaning of the whole sequence can change, while the text looks entirely normal.
So the reliable path for this task is not "letting a model recognize it", but checking against layout, numbering, and context as simultaneous constraints.
2. Line breaks and page breaks are the norm
Sequence listings frequently span pages, span columns, and are even interrupted by figure captions. If a cross-page break is treated as two independent sequences, it directly produces two erroneous records—and errors of this kind are very hard to detect downstream.
3. The numbering system may be misaligned with the body text
SEQ ID NO numbering in patents needs to be aligned across three places: body-text citations, the sequence listing, and the figures. In practice, the same number can point to different sequences in different places, or the body text can cite a number that doesn't exist. So numbering alignment must be handled as a separate step, not as a byproduct of extraction.
4. Modification information is written inconsistently
Modifications such as phosphorylation, glycosylation, and methylation are represented differently across papers: some are written inline in the sequence, some in a separate list, and some only described in prose. Normalizing directly would hide those differences—and those differences are often exactly what downstream users care about.
What We Cover
The sequence body and its attached information- Protein and nucleic acid sequence extraction and format normalization
- Continuation reconstruction across pages, columns, and figure captions
- Position-by-position alignment of similar sequences and annotation of differing positions
- Export in common formats (FASTA, etc.)
- Three-way alignment of the SEQ ID NO numbering system with body-text citations
- Normalization of modification type, position, and notation (original notation retained)
- Database cross-references (e.g., UniProt, GenBank accession numbers)
- Source location: page and line numbers in literature, patents, and figures
Short peptides, oligonucleotides, and antibody light- and heavy-chain sequences are all within scope. Antibody data usually links additionally to antigen, chain pairing, and affinity constants, and that chain can be run together with the workflow on this page.
Delivery Pipeline
Four steps in one line · fully loggedDeliverables
What you get is not just an annotation result- Structured dataset: JSON / CSV, with fields corresponding one-to-one to the Schema; sequence bodies can be exported in common formats such as FASTA
- Annotation guidelines document: field definitions, edge cases, acceptance criteria
- QC report: sampling ratio, error categories, rework records
- Provenance log: each sequence maps to its source location, annotator, and modification time
Related Cases
Real delivery scenariosExtraction and normalization of protein and nucleic acid sequences, modification information, sequence numbering, and cross-page information, covering patent and literature data in the biological sequences domain.
Sequence normalizationModification decompositionCross-page continuationFrequently Asked Questions
Questions we're often asked before working togetherHow do you handle sequences broken across pages or columns?
Rebuild the whole sequence according to continuation rules, and record the breakpoint locations and the basis for continuation in the provenance log. A cross-page break is not treated as two independent sequences, and nothing is stitched together based on length alone—continuation must have a basis in layout or numbering.
What do you do when modified residues are written in inconsistent notations?
Keep the original notation and set up a separate normalized field, recording the modification type and position. Different papers may write the same modification differently, and normalizing directly would hide those differences—and those differences are often exactly what downstream users care about.
Can the SEQ ID NO numbering in a patent fail to match the body text?
It happens. Instances where body-text citations and sequence-listing numbers are misaligned, or where the same number points to different sequences, both exist—so numbering alignment is a separate processing step, not a byproduct of extraction.
Will similar sequences get merged into one?
No. Sequence similarity and sequence identity are two different things; the determination is based on position-by-position alignment, and we do not merge things just because they look alike. Differing positions are marked separately.
What formats do you output?
Delivered as JSON / CSV, while sequence bodies can be exported in common formats such as FASTA; each record carries its original source location, modification information, and cross-reference identifiers.
Continue Reading
Related pagesThe other half besides sequences: structuring activity values together with their test conditions.
Browse all the solutions we currently offer by industry domain and specific task.
Send us a representative patent or sequence listing, and we can run a small pilot annotation batch so you can see directly how breakpoint continuation and numbering alignment are handled.