Reference
Where the data comes from
When the assistant needs a sequence or a reference it fetches it from a public repository rather than recalling it — so there is always a record you can open.
What it can reach
| Source | Good for |
|---|---|
| Addgene | Deposited plasmids. Full sequences and the depositor's map, by catalogue number. |
| iGEM Registry | Standard biological parts — promoters, RBSs, terminators, reporters — with part identifiers. |
| UniProt | Protein sequences, domain boundaries, post-translational modifications, subcellular localisation. |
| NCBI (nucleotide and protein) | Reference sequences by accession, across essentially every organism. |
| RCSB PDB | Sequences of solved structures — the right way to get the construct that actually crystallised. |
| PubMed | The literature, searchable, with full records for individual articles. |
| CrossRef | Checking that a DOI resolves to the paper it is claimed to be. |
| NCBI BLAST | Sequence similarity search against the public databases. |
| Built-in backbone catalogue | Canonical sequences for common cloning and expression vectors, held locally. |
Asking so you get the right record
- Give the identifier where you have one
- An Addgene number, a UniProt accession, a PDB ID, an NCBI accession. It resolves to exactly one record. A name resolves to a family, and you may not get the member you meant.
- Say the organism
- Gene symbols are reused across species, and the human and mouse versions of a gene are not interchangeable in your construct.
- Say the variant
- If you need a specific engineered variant, name it and expect it to be sourced. A wild-type sequence returned for a variant request is a real and common failure — and it is easy to miss because the name in the reply is the one you asked for.
Checking what comes back
A sequence
Three quick checks catch nearly everything:
| Check | Why |
|---|---|
| Open the accession | It is a link to a public database. If it does not resolve, the record is not real and nothing built on it is either. |
| Compare the length | The number of bases or residues against what the source record says. A mismatch means you have a fragment, a different isoform, or a different construct. |
| Read the first few residues | For a protein, the N-terminus is a fast fingerprint. If it does not match the published sequence, it is not the same protein. |
A citation
Fabricated references are the best-documented failure mode of language models, and biology citations are no exception. GeneLoop can check a DOI against CrossRef and tell you whether it resolves — but the habit worth keeping is simply to open anything you intend to cite.
What it cannot reach
It does not have access to your institution's subscriptions, so paywalled full text is out of reach — abstracts and metadata are not. It does not browse the open web. And it is not connected to any vendor's ordering system: it can assess whether a sequence looks manufacturable, but it cannot get you a quote or place an order.
Your own data comes first
For anything in your project, the file is the source of truth — not a published map of what that vector is supposed to be. Lab plasmids diverge from their reference sequences, often years ago and often without anyone renaming the file.
This is why cut-site searches and analyses should be run against the construct in your project rather than a catalogue entry with the same name. Bench tools.