> ## Documentation Index
> Fetch the complete documentation index at: https://docs.oneshotagent.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Predict: Alignment

> Compare biological sequences against each other or a fixed public reference, with per-hit identity, coverage and aligned regions.

Runs local sequence comparison, as the `alignment` model of [Predict](https://docs.oneshotagent.com/api-reference/predict.md). It shares the Predict request path, quote and result delivery; this page covers the `alignment` input and result only.

```json theme={null}
{
  "model": "alignment",
  "time_limit_s": 30,
  "alignment": {
    "mode": "pairwise",
    "sequence_type": "protein",
    "sequences": [
      { "id": "query", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQ" },
      { "id": "target", "sequence": "MKTAYIAKQRQISFVKSHFSRQLEERLGAIEVA" }
    ]
  }
}
```

* `mode`: `pairwise` (compare every unordered pair of 2 to 50 sequences; the earlier entry is the query, the later the target) or `search` (run 1 to 10 queries against a fixed public reference).
* `sequence_type`: `auto` (default), `protein` or `nucleotide` for the whole list. Under `auto` the type is detected and an alphabet-ambiguous protein must be stated as `protein`.
* `sequences[]`: `{ id, sequence }`, with ids 1 to 128 characters and unique.
* `time_limit_s`: 1 to 120 seconds, default 20.

Sequence limits:

* Protein: the 20 canonical letters, 20 to 10,000 residues.
* Nucleotide: A, C, G, T, U, 66 to 10,000 bases, with U read as T and a T/U mix refused.
* Combined length: at most 100,000.
* Refused: gaps, whitespace, ambiguity letters, headers, reads, quality scores and file formats.

Limits are checked before quoting and refused with `invalid_input` and a `code`: `sequence_count`, `sequence_length`, `sequence_alphabet`, `total_residues` or `mixed_sequence_types`.

`prediction.queries` holds one `{ id, hits }` entry per sequence in input order. In `pairwise` only later entries are targets; no match is an empty list. Each hit carries:

* `target_id`, `identity`, `coverage`, `e_value`, `bit_score`
* `aligned_region`: 1-based and inclusive, on the original sequences
* `reference_release`: `2025_01` for `search`, `null` for `pairwise`

`meta` reports the resolved type and the comparison counts; `checks` reports the input and consistency checks. On a timeout only completed comparisons are returned.

`search` runs against UniProtKB reviewed (Swiss-Prot) release `2025_01`, and the result's `reference_attribution` field carries the attribution its licence requires:

> UniProtKB reviewed (Swiss-Prot) release 2025\_01, distributed under CC BY 4.0. Source: [https://www.uniprot.org/](https://www.uniprot.org/).

Some requests are declined: HTTP 422 `{ "error": "request_refused", "message": "This request cannot be processed." }`, or HTTP 503 with code `service_unavailable` and the same message when the service is temporarily unavailable. A refused request is not charged.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.