---
title: AlphaFold inputs
description: Give Ivy one input, a UniProt accession like P00533 — entry names, gene symbols and PDB IDs do not resolve, and isoforms are canonical only.
---

You provide one thing: a UniProt accession for the protein you want. Getting that accession right is what most affects the result — an entry name, gene symbol, or PDB ID won't resolve directly, and whether a structure comes back at all depends on the protein being covered by the AlphaFold Database.

## Inputs

You ask for an AlphaFold structure in chat; the lookup needs one thing, a **UniProt accession**. There is no form or job for it.

## UniProt accession (required)

A UniProt accession, e.g. `P00533` (human EGFR), `P68871` (hemoglobin subunit beta), `Q9Y6K9` (NEMO).

### Accepted formats

| Form | Example | Accepted |
|---|---|---|
| UniProt accession | `P00533` | ✓ |
| UniProt entry name | `EGFR_HUMAN` | ⚠︎ Ivy resolves it to an accession first with a UniProt search |
| PDB ID | `6LU7` | ✗ Ask for the [PDB entry](/docs/pdb) instead |
| Raw sequence | `MKTAY...` | ✗ Ask for an [ESMFold prediction](/docs/esmfold) instead |
| Gene symbol | `EGFR` | ✗ Ivy resolves it with a UniProt search first |

If you have a gene name or protein name rather than an accession, just ask in chat — Ivy runs a UniProt search, then the AlphaFold fetch, automatically.

## How to find a UniProt accession

- **By name**: in chat, *"Find UniProt entry for EGFR"* → runs a UniProt search
- **By BLAST**: provide a sequence and BLAST returns the closest UniProt hits
- **Manually**: search at [uniprot.org](https://www.uniprot.org/uniprotkb)

## Database coverage limits

Not every UniProt entry has an AlphaFold structure. The database covers:

- All reviewed (Swiss-Prot) human entries
- Most reviewed entries from major model organisms
- A large subset of TrEMBL (unreviewed) entries for covered proteomes

If the fetch returns "not found":

- Ask Ivy to [predict it with ESMFold](/docs/esmfold) from the UniProt sequence instead
- Or ask in one message to try AlphaFold first and fall back to ESMFold; the [orchestrator](/docs/how-orchestration-works) chains both

## What the AlphaFold fetch does NOT take

- **Custom sequences** — the fetch only looks up pre-computed structures. For novel or mutated sequences, ask for an [ESMFold prediction](/docs/esmfold).
- **Multi-chain complexes** — a UniProt entry in the AlphaFold DB is a single chain, modelled alone, with no partners, ligands, ions or cofactors. For multimers, use a co-folding model ([Boltz-2](/docs/boltz2), [Chai-1](/docs/chai)) or AlphaFold-Multimer via a different workflow. Published AlphaFold figures on complexes almost always describe AlphaFold-Multimer, which is not what this tool returns — see [translatability](/docs/alphafold/translatability#on-complexes-and-the-version-trap).
- **Specific isoforms beyond canonical** — the fetch returns the database's `F1` entry for the accession. Non-canonical isoforms need an ESMFold prediction on the isoform sequence.

## Read next