scigantic-alphafold 0.1.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (28) hide show
  1. scigantic_alphafold-0.1.0/CHANGELOG.md +24 -0
  2. scigantic_alphafold-0.1.0/LICENSE +16 -0
  3. scigantic_alphafold-0.1.0/MANIFEST.in +6 -0
  4. scigantic_alphafold-0.1.0/PKG-INFO +226 -0
  5. scigantic_alphafold-0.1.0/README.md +199 -0
  6. scigantic_alphafold-0.1.0/pyproject.toml +59 -0
  7. scigantic_alphafold-0.1.0/setup.cfg +4 -0
  8. scigantic_alphafold-0.1.0/src/scigantic_alphafold/__init__.py +57 -0
  9. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_api.py +218 -0
  10. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_build.py +1639 -0
  11. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_errors.py +77 -0
  12. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_fetch.py +304 -0
  13. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_index.py +819 -0
  14. scigantic_alphafold-0.1.0/src/scigantic_alphafold/_version.py +2 -0
  15. scigantic_alphafold-0.1.0/src/scigantic_alphafold/cli.py +160 -0
  16. scigantic_alphafold-0.1.0/src/scigantic_alphafold/py.typed +0 -0
  17. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/PKG-INFO +226 -0
  18. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/SOURCES.txt +26 -0
  19. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/dependency_links.txt +1 -0
  20. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/entry_points.txt +2 -0
  21. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/requires.txt +4 -0
  22. scigantic_alphafold-0.1.0/src/scigantic_alphafold.egg-info/top_level.txt +1 -0
  23. scigantic_alphafold-0.1.0/tests/conftest.py +309 -0
  24. scigantic_alphafold-0.1.0/tests/test_api.py +443 -0
  25. scigantic_alphafold-0.1.0/tests/test_build.py +1452 -0
  26. scigantic_alphafold-0.1.0/tests/test_cli.py +415 -0
  27. scigantic_alphafold-0.1.0/tests/test_fetch.py +548 -0
  28. scigantic_alphafold-0.1.0/tests/test_index.py +852 -0
@@ -0,0 +1,24 @@
1
+ # Changelog
2
+
3
+ All notable changes to this project are recorded here. The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
4
+
5
+ ## 0.1.0 - 2026-11-20
6
+
7
+ First release.
8
+
9
+ ### Added
10
+
11
+ - `fetch`, `fetch_bytes`, `download` and `fetch_many`: read single AlphaFold DB v6 structures (mmCIF or PDB) with one HTTP range request each against EBI's public bulk tars at `https://ftp.ebi.ac.uk/pub/databases/alphafold/v6/`.
12
+ - `locate`, `has`, `entries`, `tars` and `index_info`: answer from the offset index without a request to EBI.
13
+ - The offset index `index-v6.sqlite` (schema version 1), downloaded on first use from this repository's `index-v6` release and cached in `~/.cache/scigantic-alphafold`. `SCIGANTIC_ALPHAFOLD_INDEX` and `SCIGANTIC_ALPHAFOLD_CACHE` override the source and the cache directory. `refresh_index()` downloads it again.
14
+ - `NotInBulkSetError` for accessions outside the bulk set (Swiss-Prot plus 46 proteome tars) and `StaleIndexError` when a tar on the server no longer matches the index.
15
+ - `metadata()`: passthrough of EBI's prediction API, `https://alphafold.ebi.ac.uk/api/prediction/<qualifier>`. Returns `[]` for an accession with no prediction, raises `InvalidIdError` for input that is not an accession or model id, and raises `ApiError` for HTTP errors.
16
+ - `proteomes()`: passthrough of EBI's `download_metadata.json`, with `url`, `tax_id` and `reference_proteome` added to each row.
17
+ - The `scigantic-alphafold` command: `get`, `locate`, `info`, `refresh`, and the index builders `build-part` and `merge`.
18
+ - `examples/quickstart.ipynb`.
19
+
20
+ ### Notes
21
+
22
+ - Coverage is the v6 bulk set only, coordinates only. There are no PAE files and no MSAs.
23
+ - No runtime dependencies. Python 3.10 or newer.
24
+ - Every request sends the User-Agent `scigantic-alphafold/<version> (+https://github.com/Scigantic/scigantic-alphafold)`.
@@ -0,0 +1,16 @@
1
+ MIT No Attribution
2
+
3
+ Copyright 2026 Scigantic
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy of this
6
+ software and associated documentation files (the "Software"), to deal in the Software
7
+ without restriction, including without limitation the rights to use, copy, modify,
8
+ merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
9
+ permit persons to whom the Software is furnished to do so.
10
+
11
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
12
+ INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
13
+ PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
14
+ HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
15
+ CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE
16
+ OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
@@ -0,0 +1,6 @@
1
+ include LICENSE README.md CHANGELOG.md
2
+ # The tests need their shared fixtures to run from an unpacked sdist.
3
+ include tests/conftest.py
4
+ recursive-include tests *.py
5
+ prune build
6
+ prune dist
@@ -0,0 +1,226 @@
1
+ Metadata-Version: 2.4
2
+ Name: scigantic-alphafold
3
+ Version: 0.1.0
4
+ Summary: Read single AlphaFold DB v6 structures with one HTTP range request against EBI's public bulk tars, using a prebuilt byte-offset index. No mirror, no account, no key, no runtime dependencies.
5
+ Author: Scigantic
6
+ License-Expression: MIT-0
7
+ Project-URL: Homepage, https://scigantic.com
8
+ Project-URL: Repository, https://github.com/Scigantic/scigantic-alphafold
9
+ Project-URL: Issues, https://github.com/Scigantic/scigantic-alphafold/issues
10
+ Keywords: alphafold,afdb,protein-structure,mmcif,pdb,uniprot,structural-biology,range-request
11
+ Classifier: Programming Language :: Python :: 3
12
+ Classifier: Programming Language :: Python :: 3.10
13
+ Classifier: Programming Language :: Python :: 3.11
14
+ Classifier: Programming Language :: Python :: 3.12
15
+ Classifier: Programming Language :: Python :: 3.13
16
+ Classifier: Programming Language :: Python :: 3.14
17
+ Classifier: Typing :: Typed
18
+ Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
19
+ Classifier: Topic :: Scientific/Engineering :: Chemistry
20
+ Requires-Python: >=3.10
21
+ Description-Content-Type: text/markdown
22
+ License-File: LICENSE
23
+ Provides-Extra: dev
24
+ Requires-Dist: pytest>=7; extra == "dev"
25
+ Requires-Dist: mypy>=1.10; extra == "dev"
26
+ Dynamic: license-file
27
+
28
+ <h1 align="center">scigantic-alphafold</h1>
29
+
30
+ <p align="center">
31
+ <a href="https://github.com/Scigantic/scigantic-alphafold/actions/workflows/ci.yml">
32
+ <img alt="CI" src="https://github.com/Scigantic/scigantic-alphafold/actions/workflows/ci.yml/badge.svg" /></a>
33
+ <a href="https://pypi.org/project/scigantic-alphafold/">
34
+ <img alt="PyPI" src="https://img.shields.io/pypi/v/scigantic-alphafold" /></a>
35
+ <a href="https://pypi.org/project/scigantic-alphafold/">
36
+ <img alt="PyPI - Python Version" src="https://img.shields.io/pypi/pyversions/scigantic-alphafold" /></a>
37
+ <a href="https://github.com/Scigantic/scigantic-alphafold/blob/main/LICENSE">
38
+ <img alt="License" src="https://img.shields.io/github/license/Scigantic/scigantic-alphafold" /></a>
39
+ </p>
40
+
41
+ Read a single [AlphaFold DB](https://alphafold.ebi.ac.uk/) v6 structure with one HTTP range request against EBI's public bulk tars, using a prebuilt byte-offset index. No mirror, no account, no key, and no multi-gigabyte tar download to get one protein.
42
+
43
+ ```python
44
+ import scigantic_alphafold as af
45
+
46
+ cif = af.fetch("P69905") # hemoglobin alpha, mmCIF text
47
+ pdb = af.fetch("P69905", fmt="pdb") # the same model as PDB text
48
+ path = af.download("P69905", "structures") # writes the .cif, returns a pathlib.Path
49
+ many = af.fetch_many(["P0A7G6", "P0A6F5", "P00350"]) # dict of id to text, 4 connections by default
50
+
51
+ af.has("P69905") # True: the accession is in the bulk set
52
+ af.locate("P69905") # [Member(entry_id, fmt, tar, url, offset, length), ...]
53
+ ids = af.entries(proteome="UP000000625") # iterator over the entry ids of the E. coli tar
54
+
55
+ af.metadata("P69905") # EBI prediction API records, no index needed
56
+ af.proteomes() # EBI's list of bulk tars, with url and tax_id
57
+ ```
58
+
59
+ A longer walkthrough, including per-residue pLDDT from the B-factor column, is in [examples/quickstart.ipynb](https://github.com/Scigantic/scigantic-alphafold/blob/main/examples/quickstart.ipynb).
60
+
61
+ **Coverage is limited to the v6 bulk set: Swiss-Prot plus 46 proteomes, coordinates only.** Most of AlphaFold DB is outside that set and cannot be read with this library. See [What is covered](#what-is-covered) before you depend on it.
62
+
63
+ ## Installation
64
+
65
+ ```console
66
+ $ pip install scigantic-alphafold
67
+ ```
68
+
69
+ Python 3.10 or newer. There are no runtime dependencies; the package uses only the standard library.
70
+
71
+ ## Why this exists
72
+
73
+ EBI publishes AlphaFold DB v6 in bulk as 48 tar files at `https://ftp.ebi.ac.uk/pub/databases/alphafold/v6/`, 153.2 GB in total. The smallest is 173 MB (*Helicobacter pylori*), the human proteome is 5.2 GB and the Swiss-Prot mmCIF tar is 40.1 GB. A tar has no table of contents, so finding one protein in it normally means downloading the whole file.
74
+
75
+ Inside each tar every structure is its own small gzip file, and EBI's server honors HTTP `Range` requests. If you know the byte offset and length of a member, you can read that member alone. This package ships that knowledge as an index and does the range read.
76
+
77
+ On 2026-11-20 we observed that EBI's per-file URLs (`https://alphafold.ebi.ac.uk/files/AF-<accession>-F1-model_v6.cif`) returned HTTP 403 to scripted requests, while the prediction API and the FTP area answered normally. We know of no announcement about this and do not know whether it is intended or permanent. This library does not use the per-file endpoint. It reads only the bulk files that EBI publishes for download.
78
+
79
+ ## What is covered
80
+
81
+ The index covers the 48 tars listed in EBI's [`download_metadata.json`](https://ftp.ebi.ac.uk/pub/databases/alphafold/download_metadata.json) (counts and sizes checked on 2026-11-20):
82
+
83
+ | set | tars | structures | size |
84
+ |---|---|---|---|
85
+ | Swiss-Prot, mmCIF | 1 | 550,122 | 40.1 GB |
86
+ | Swiss-Prot, PDB | 1 | 550,122 | 28.6 GB |
87
+ | Model organism proteomes | 16 | 328,602 | 49.9 GB |
88
+ | Global health proteomes | 30 | 227,244 | 34.7 GB |
89
+
90
+ The proteome tars hold both formats for every entry. The 16 model organisms include human (UP000005640), mouse, rat, zebrafish, fruit fly, *C. elegans*, yeast, *E. coli* and *Arabidopsis*. `af.proteomes()` returns the full list with species, taxonomy id, reference proteome, structure count, size and URL. A protein can be in both Swiss-Prot and a proteome tar, so the counts above overlap.
91
+
92
+ What is not covered:
93
+
94
+ - **Any accession outside those tars.** AlphaFold DB holds predictions for far more UniProt entries than the bulk set. For an accession that is in no v6 bulk tar, `fetch`, `fetch_bytes` and `download` raise `NotInBulkSetError` (a `LookupError`). `af.locate(accession)` returns an empty list and `af.has(accession)` returns `False`, both without raising.
95
+ - **PAE.** The bulk tars hold coordinates only. There are no predicted aligned error files.
96
+ - **MSAs.** There are no alignment files.
97
+ - **Other files EBI links from an entry page**, such as the per-residue confidence JSON, binary CIF and AlphaMissense annotations. Per-residue pLDDT is still available, because it is stored in the B-factor column of the coordinates.
98
+ - **Older versions.** The index describes v6 only.
99
+
100
+ `af.metadata()` is the exception to the coverage limit. It calls EBI's prediction API, which answers for all of AlphaFold DB, so a record from `metadata()` does not mean the structure is in the bulk set.
101
+
102
+ According to EBI's [download page](https://alphafold.ebi.ac.uk/download), proteins longer than 2,700 residues are predicted as overlapping 1,400-residue fragments (`F1`, `F2`, ...), and those fragments are in the human proteome tar. An entry id such as `AF-Q8WZ42-F2` names one fragment.
103
+
104
+ ## Public API
105
+
106
+ | function | returns |
107
+ |---|---|
108
+ | `fetch(id, fmt=None)` | the decompressed text of one structure (`str`) |
109
+ | `fetch_bytes(id, fmt=None)` | the gzip bytes exactly as stored in the tar |
110
+ | `download(id, out_dir, fmt=None, decompress=True)` | `pathlib.Path` of the file written |
111
+ | `fetch_many(ids, fmt=None, workers=4, on_error="raise")` | `dict[str, str]` keyed by the ids as given, fetched on a thread pool |
112
+ | `locate(id, fmt=None)` | `list[Member]`; a `Member` has `entry_id`, `fmt`, `tar`, `url`, `offset`, `length` |
113
+ | `has(id)` | `bool`, answered from the index without a request to EBI |
114
+ | `entries(proteome=..., fmt="cif")`, `entries(tar=..., fmt="cif")` | iterator over the entry ids of one tar |
115
+ | `tars()` | `list[Tar]`, the tars the index covers |
116
+ | `index_info()` | `dict` describing the index in use |
117
+ | `refresh_index()` | `pathlib.Path` of the freshly downloaded index |
118
+ | `metadata(qualifier, timeout=30.0)` | `list[dict]` from EBI's prediction API |
119
+ | `proteomes(timeout=30.0)` | `list[dict]` from EBI's `download_metadata.json` |
120
+
121
+ `id` is a UniProt accession (`P69905`) or an AlphaFold DB entry id (`AF-P69905-F1`). `fmt` is `"cif"` or `"pdb"`; left as `None` it means mmCIF. A member file name such as `AF-P69905-F1-model_v6.pdb.gz` is also accepted as an id and selects its own format; a file name for another version, or one whose extension conflicts with an explicit `fmt`, raises `InvalidIdError`.
122
+
123
+ How an id resolves:
124
+
125
+ - A bare accession resolves to `AF-<accession>-F1` when that is its only fragment.
126
+ - An accession that is split into several fragments (`F1`, `F2`, ...) raises `AmbiguousEntryError` listing the entry ids, because `F1` alone is only the first part of the protein. Pass the entry id of the fragment you want. `locate()` returns all of them.
127
+ - An isoform is its own accession (`P12345-2`).
128
+ - When the same entry is in two tars (a Swiss-Prot protein that is also in a proteome tar), the tar whose name sorts first is read, which in practice is the proteome tar. `locate()` lists that copy first.
129
+
130
+ `fetch_many` attempts every id. If any fail it raises `FetchManyError` after the rest have finished, with the successes in `.results` and the failures in `.errors`. Pass `on_error="skip"` to get only the successes back.
131
+
132
+ Errors:
133
+
134
+ - `NotInBulkSetError` (a `LookupError`): the accession is in no v6 bulk tar.
135
+ - `AmbiguousEntryError`: a bare accession has several fragments, or only isoform entries.
136
+ - `StaleIndexError` (a `RuntimeError`): EBI replaced a tar after the index was built. Call `refresh_index()`.
137
+ - `FetchError`: EBI answered a range request with an HTTP error. It carries `status`.
138
+ - `FetchManyError`: one or more ids in `fetch_many` failed. It carries `results` and `errors`.
139
+ - `IndexUnavailableError`, `IndexSchemaError`: the index could not be downloaded, or was built for a newer version of this library.
140
+ - `InvalidIdError` (a `ValueError`): the id is not an accession or entry id.
141
+ - `ApiError` (a `RuntimeError`): `metadata()` or `proteomes()` got an HTTP error, a timeout or an unexpected body. It carries `status` and `url`.
142
+
143
+ ### `metadata()` and `proteomes()`
144
+
145
+ Both are thin passthroughs. They need no index, and they return what EBI sends.
146
+
147
+ `metadata("P69905")` calls `https://alphafold.ebi.ac.uk/api/prediction/P69905` and returns a list of records with keys such as `entryId`, `uniprotAccession`, `gene`, `taxId`, `organismScientificName`, `sequence`, `globalMetricValue` (mean pLDDT) and `latestVersion`. The API answers a well-formed accession that has no prediction with HTTP 404 and an empty JSON object; `metadata()` returns `[]` for that case. A value that is not made of letters, digits, `-` and `_` raises `InvalidIdError` (a `ValueError`) before any request is sent. An identifier the server rejects (HTTP 400) raises `ApiError` with the server's message.
148
+
149
+ `proteomes()` returns one dict per bulk tar with EBI's fields (`archive_name`, `type`, `num_predicted_structures`, `size_bytes`, and for proteome tars `species`, `common_name` and `reference_proteome`) plus `url`, the location of the tar, and `tax_id`, parsed from names such as `UP000000625_83333_ECOLI_v6.tar`. `tax_id` and `reference_proteome` are `None` for the two Swiss-Prot rows.
150
+
151
+ Both functions retry timeouts and 5xx responses twice with backoff, and never retry a 4xx.
152
+
153
+ ## How it works
154
+
155
+ 1. **The index.** One SQLite file, `index-v6.sqlite`, has a row per structure file: entry id, format, tar, byte offset and length of the gzip member. It also records each tar's size, `Last-Modified` and `ETag` at build time.
156
+ 2. **Where the index is hosted.** It is a release asset of this repository: `https://github.com/Scigantic/scigantic-alphafold/releases/download/index-v6/index-v6.sqlite.gz`. EBI does not host or maintain it.
157
+ 3. **First use.** The first call that needs the index downloads it to the cache directory, `~/.cache/scigantic-alphafold` by default. The download goes to a temporary name and is renamed into place, so an interrupted download does not leave a partial index. The download is checked against the SHA-256 published next to it in `index-v6.json`; a mismatch is refused, which can happen for a moment while the index is being republished. Later calls open the cached file. `import scigantic_alphafold` alone does not download anything.
158
+ 4. **One range read.** `fetch` looks up the entry, sends one `GET` with a `Range` header to the tar on `ftp.ebi.ac.uk`, and gunzips the bytes that come back. A member is small: the mmCIF of a 217-residue *H. pylori* protein (`AF-G5EB01-F1`) is 47,367 bytes gzipped, read out of a 173 MB tar.
159
+ 5. **Staleness check.** If the `Last-Modified` or `ETag` of the response differs from what the index recorded, the bytes do not gunzip, or the file name recorded in the gzip header is not the entry that was asked for, the offsets can no longer be trusted and the call raises `StaleIndexError`. `af.refresh_index()` downloads the current index again.
160
+
161
+ These environment variables change where the index comes from and what it may point at:
162
+
163
+ | variable | meaning |
164
+ |---|---|
165
+ | `SCIGANTIC_ALPHAFOLD_INDEX` | a local path or a URL of an index to use instead of the default, as `.sqlite` or `.sqlite.gz` |
166
+ | `SCIGANTIC_ALPHAFOLD_CACHE` | the cache directory, instead of `~/.cache/scigantic-alphafold` |
167
+ | `SCIGANTIC_ALPHAFOLD_ALLOW_ANY_HOST` | set to `1` to accept an index whose tars are not on `https://ftp.ebi.ac.uk`, for a private mirror or for tests |
168
+
169
+ By default the library only sends range requests to `https://ftp.ebi.ac.uk`, whatever the index says, so a wrong or tampered index cannot point it at another host.
170
+
171
+ `af.index_info()` reports which index is in use. The index is built by reading every tar once and recording where each member starts; [builders/README.md](https://github.com/Scigantic/scigantic-alphafold/blob/main/builders/README.md) describes the build and how to reproduce it.
172
+
173
+ ## Being a polite client of EBI
174
+
175
+ The FTP area is a shared public service that EMBL-EBI runs at its own cost. This library is meant for reading some structures, not for copying the database.
176
+
177
+ - Every request carries an honest `User-Agent`: `scigantic-alphafold/<version> (+https://github.com/Scigantic/scigantic-alphafold)`. Do not change it to look like a browser.
178
+ - `fetch_many` uses 4 connections by default. Keep `workers` near that, and do not run several processes in parallel against the FTP.
179
+ - Requests are retried with backoff only after timeouts and 5xx responses.
180
+ - If you need a whole proteome or all of Swiss-Prot, do not loop over `fetch`. Download EBI's tar once: `af.proteomes()` lists the URLs. For large transfers EBI recommends [Globus](https://app.globus.org/file-manager?origin_id=47772002-3e5b-4fd3-b97c-18cee38d6df2&origin_path=%2Fpub%2Fdatabases%2Falphafold%2F) (endpoint "EMBL-EBI Public Data", path `/pub/databases/alphafold/`), as described in the [FTP README](https://ftp.ebi.ac.uk/pub/databases/alphafold/README.txt).
181
+ - If EBI limits or closes range access to the tars, this library will stop working, and it will not try to get around that.
182
+
183
+ ## Command line
184
+
185
+ `scigantic-alphafold` wraps the same functions.
186
+
187
+ ```console
188
+ $ scigantic-alphafold get P69905 # writes the mmCIF file to the current directory
189
+ $ scigantic-alphafold get P69905 P68871 -f pdb -o structures
190
+ $ scigantic-alphafold get P69905 --stdout | head
191
+ $ scigantic-alphafold locate P69905 # tar, offset and length
192
+ $ scigantic-alphafold info # the index in use
193
+ $ scigantic-alphafold refresh # download the index again
194
+ ```
195
+
196
+ Two more subcommands, `build-part <tar>` and `merge <parts dir> -o index-v6.sqlite`, build the index; `merge --only <tar>...` builds a partial index from the tars you streamed. They are documented in [builders/README.md](https://github.com/Scigantic/scigantic-alphafold/blob/main/builders/README.md).
197
+
198
+ ## Data license and citation
199
+
200
+ This project is not affiliated with, endorsed by or supported by EMBL-EBI or Google DeepMind. It redistributes no AlphaFold DB data: the structures are read from EBI's servers at the time of the call, and the index holds only file names, byte offsets and lengths.
201
+
202
+ The [AlphaFold DB site](https://alphafold.ebi.ac.uk/) states: "Data is available for academic and commercial use, under a [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) licence." It also states: "AlphaFold Data Copyright (2022) DeepMind Technologies Limited." (both read on 2026-11-20). CC BY 4.0 requires attribution. The site's disclaimer says the data "is for theoretical modelling only" and has "not been validated for, and are not approved for, any clinical use"; read it before relying on a structure. Use of AlphaFold DB is also subject to the [EMBL-EBI Terms of Use](https://www.ebi.ac.uk/about/terms-of-use).
203
+
204
+ If you use structures obtained through this library, cite AlphaFold DB and AlphaFold, not this package. The papers the site asked users to cite on 2026-11-20:
205
+
206
+ - Bertoni D. et al. *AlphaFold Protein Structure Database 2025: a redesigned interface and updated structural coverage*. Nucleic Acids Research (2025). [academic.oup.com/nar/article/54/D1/D358/8340156](https://academic.oup.com/nar/article/54/D1/D358/8340156). Listed on the home page for use of the resource.
207
+ - Fleming J. et al. *AlphaFold Protein Structure Database and 3D-Beacons: New Data and Capabilities*. Journal of Molecular Biology (2025). [sciencedirect.com/science/article/pii/S0022283625000336](https://www.sciencedirect.com/science/article/pii/S0022283625000336). Listed on the download page for use of a prediction.
208
+ - Jumper J. et al. *Highly accurate protein structure prediction with AlphaFold*. Nature (2021). [doi:10.1038/s41586-021-03819-2](https://doi.org/10.1038/s41586-021-03819-2). Listed on the download page for predictions from the UniProt dataset.
209
+
210
+ The site's guidance can change. Check [alphafold.ebi.ac.uk](https://alphafold.ebi.ac.uk/) for the current text before you publish.
211
+
212
+ This package's code is MIT-0. That does not extend to the data.
213
+
214
+ ## Testing
215
+
216
+ ```console
217
+ $ pip install -e ".[dev]"
218
+ $ pytest
219
+ $ mypy --strict src/scigantic_alphafold
220
+ ```
221
+
222
+ The default run is offline: tests use local files and an HTTP server on localhost. Tests marked `live` send a few small requests to EBI and are skipped unless `SCIGANTIC_ALPHAFOLD_LIVE=1` is set.
223
+
224
+ ## License
225
+
226
+ MIT-0 for the code. See [Data license and citation](#data-license-and-citation) for the data.
@@ -0,0 +1,199 @@
1
+ <h1 align="center">scigantic-alphafold</h1>
2
+
3
+ <p align="center">
4
+ <a href="https://github.com/Scigantic/scigantic-alphafold/actions/workflows/ci.yml">
5
+ <img alt="CI" src="https://github.com/Scigantic/scigantic-alphafold/actions/workflows/ci.yml/badge.svg" /></a>
6
+ <a href="https://pypi.org/project/scigantic-alphafold/">
7
+ <img alt="PyPI" src="https://img.shields.io/pypi/v/scigantic-alphafold" /></a>
8
+ <a href="https://pypi.org/project/scigantic-alphafold/">
9
+ <img alt="PyPI - Python Version" src="https://img.shields.io/pypi/pyversions/scigantic-alphafold" /></a>
10
+ <a href="https://github.com/Scigantic/scigantic-alphafold/blob/main/LICENSE">
11
+ <img alt="License" src="https://img.shields.io/github/license/Scigantic/scigantic-alphafold" /></a>
12
+ </p>
13
+
14
+ Read a single [AlphaFold DB](https://alphafold.ebi.ac.uk/) v6 structure with one HTTP range request against EBI's public bulk tars, using a prebuilt byte-offset index. No mirror, no account, no key, and no multi-gigabyte tar download to get one protein.
15
+
16
+ ```python
17
+ import scigantic_alphafold as af
18
+
19
+ cif = af.fetch("P69905") # hemoglobin alpha, mmCIF text
20
+ pdb = af.fetch("P69905", fmt="pdb") # the same model as PDB text
21
+ path = af.download("P69905", "structures") # writes the .cif, returns a pathlib.Path
22
+ many = af.fetch_many(["P0A7G6", "P0A6F5", "P00350"]) # dict of id to text, 4 connections by default
23
+
24
+ af.has("P69905") # True: the accession is in the bulk set
25
+ af.locate("P69905") # [Member(entry_id, fmt, tar, url, offset, length), ...]
26
+ ids = af.entries(proteome="UP000000625") # iterator over the entry ids of the E. coli tar
27
+
28
+ af.metadata("P69905") # EBI prediction API records, no index needed
29
+ af.proteomes() # EBI's list of bulk tars, with url and tax_id
30
+ ```
31
+
32
+ A longer walkthrough, including per-residue pLDDT from the B-factor column, is in [examples/quickstart.ipynb](https://github.com/Scigantic/scigantic-alphafold/blob/main/examples/quickstart.ipynb).
33
+
34
+ **Coverage is limited to the v6 bulk set: Swiss-Prot plus 46 proteomes, coordinates only.** Most of AlphaFold DB is outside that set and cannot be read with this library. See [What is covered](#what-is-covered) before you depend on it.
35
+
36
+ ## Installation
37
+
38
+ ```console
39
+ $ pip install scigantic-alphafold
40
+ ```
41
+
42
+ Python 3.10 or newer. There are no runtime dependencies; the package uses only the standard library.
43
+
44
+ ## Why this exists
45
+
46
+ EBI publishes AlphaFold DB v6 in bulk as 48 tar files at `https://ftp.ebi.ac.uk/pub/databases/alphafold/v6/`, 153.2 GB in total. The smallest is 173 MB (*Helicobacter pylori*), the human proteome is 5.2 GB and the Swiss-Prot mmCIF tar is 40.1 GB. A tar has no table of contents, so finding one protein in it normally means downloading the whole file.
47
+
48
+ Inside each tar every structure is its own small gzip file, and EBI's server honors HTTP `Range` requests. If you know the byte offset and length of a member, you can read that member alone. This package ships that knowledge as an index and does the range read.
49
+
50
+ On 2026-11-20 we observed that EBI's per-file URLs (`https://alphafold.ebi.ac.uk/files/AF-<accession>-F1-model_v6.cif`) returned HTTP 403 to scripted requests, while the prediction API and the FTP area answered normally. We know of no announcement about this and do not know whether it is intended or permanent. This library does not use the per-file endpoint. It reads only the bulk files that EBI publishes for download.
51
+
52
+ ## What is covered
53
+
54
+ The index covers the 48 tars listed in EBI's [`download_metadata.json`](https://ftp.ebi.ac.uk/pub/databases/alphafold/download_metadata.json) (counts and sizes checked on 2026-11-20):
55
+
56
+ | set | tars | structures | size |
57
+ |---|---|---|---|
58
+ | Swiss-Prot, mmCIF | 1 | 550,122 | 40.1 GB |
59
+ | Swiss-Prot, PDB | 1 | 550,122 | 28.6 GB |
60
+ | Model organism proteomes | 16 | 328,602 | 49.9 GB |
61
+ | Global health proteomes | 30 | 227,244 | 34.7 GB |
62
+
63
+ The proteome tars hold both formats for every entry. The 16 model organisms include human (UP000005640), mouse, rat, zebrafish, fruit fly, *C. elegans*, yeast, *E. coli* and *Arabidopsis*. `af.proteomes()` returns the full list with species, taxonomy id, reference proteome, structure count, size and URL. A protein can be in both Swiss-Prot and a proteome tar, so the counts above overlap.
64
+
65
+ What is not covered:
66
+
67
+ - **Any accession outside those tars.** AlphaFold DB holds predictions for far more UniProt entries than the bulk set. For an accession that is in no v6 bulk tar, `fetch`, `fetch_bytes` and `download` raise `NotInBulkSetError` (a `LookupError`). `af.locate(accession)` returns an empty list and `af.has(accession)` returns `False`, both without raising.
68
+ - **PAE.** The bulk tars hold coordinates only. There are no predicted aligned error files.
69
+ - **MSAs.** There are no alignment files.
70
+ - **Other files EBI links from an entry page**, such as the per-residue confidence JSON, binary CIF and AlphaMissense annotations. Per-residue pLDDT is still available, because it is stored in the B-factor column of the coordinates.
71
+ - **Older versions.** The index describes v6 only.
72
+
73
+ `af.metadata()` is the exception to the coverage limit. It calls EBI's prediction API, which answers for all of AlphaFold DB, so a record from `metadata()` does not mean the structure is in the bulk set.
74
+
75
+ According to EBI's [download page](https://alphafold.ebi.ac.uk/download), proteins longer than 2,700 residues are predicted as overlapping 1,400-residue fragments (`F1`, `F2`, ...), and those fragments are in the human proteome tar. An entry id such as `AF-Q8WZ42-F2` names one fragment.
76
+
77
+ ## Public API
78
+
79
+ | function | returns |
80
+ |---|---|
81
+ | `fetch(id, fmt=None)` | the decompressed text of one structure (`str`) |
82
+ | `fetch_bytes(id, fmt=None)` | the gzip bytes exactly as stored in the tar |
83
+ | `download(id, out_dir, fmt=None, decompress=True)` | `pathlib.Path` of the file written |
84
+ | `fetch_many(ids, fmt=None, workers=4, on_error="raise")` | `dict[str, str]` keyed by the ids as given, fetched on a thread pool |
85
+ | `locate(id, fmt=None)` | `list[Member]`; a `Member` has `entry_id`, `fmt`, `tar`, `url`, `offset`, `length` |
86
+ | `has(id)` | `bool`, answered from the index without a request to EBI |
87
+ | `entries(proteome=..., fmt="cif")`, `entries(tar=..., fmt="cif")` | iterator over the entry ids of one tar |
88
+ | `tars()` | `list[Tar]`, the tars the index covers |
89
+ | `index_info()` | `dict` describing the index in use |
90
+ | `refresh_index()` | `pathlib.Path` of the freshly downloaded index |
91
+ | `metadata(qualifier, timeout=30.0)` | `list[dict]` from EBI's prediction API |
92
+ | `proteomes(timeout=30.0)` | `list[dict]` from EBI's `download_metadata.json` |
93
+
94
+ `id` is a UniProt accession (`P69905`) or an AlphaFold DB entry id (`AF-P69905-F1`). `fmt` is `"cif"` or `"pdb"`; left as `None` it means mmCIF. A member file name such as `AF-P69905-F1-model_v6.pdb.gz` is also accepted as an id and selects its own format; a file name for another version, or one whose extension conflicts with an explicit `fmt`, raises `InvalidIdError`.
95
+
96
+ How an id resolves:
97
+
98
+ - A bare accession resolves to `AF-<accession>-F1` when that is its only fragment.
99
+ - An accession that is split into several fragments (`F1`, `F2`, ...) raises `AmbiguousEntryError` listing the entry ids, because `F1` alone is only the first part of the protein. Pass the entry id of the fragment you want. `locate()` returns all of them.
100
+ - An isoform is its own accession (`P12345-2`).
101
+ - When the same entry is in two tars (a Swiss-Prot protein that is also in a proteome tar), the tar whose name sorts first is read, which in practice is the proteome tar. `locate()` lists that copy first.
102
+
103
+ `fetch_many` attempts every id. If any fail it raises `FetchManyError` after the rest have finished, with the successes in `.results` and the failures in `.errors`. Pass `on_error="skip"` to get only the successes back.
104
+
105
+ Errors:
106
+
107
+ - `NotInBulkSetError` (a `LookupError`): the accession is in no v6 bulk tar.
108
+ - `AmbiguousEntryError`: a bare accession has several fragments, or only isoform entries.
109
+ - `StaleIndexError` (a `RuntimeError`): EBI replaced a tar after the index was built. Call `refresh_index()`.
110
+ - `FetchError`: EBI answered a range request with an HTTP error. It carries `status`.
111
+ - `FetchManyError`: one or more ids in `fetch_many` failed. It carries `results` and `errors`.
112
+ - `IndexUnavailableError`, `IndexSchemaError`: the index could not be downloaded, or was built for a newer version of this library.
113
+ - `InvalidIdError` (a `ValueError`): the id is not an accession or entry id.
114
+ - `ApiError` (a `RuntimeError`): `metadata()` or `proteomes()` got an HTTP error, a timeout or an unexpected body. It carries `status` and `url`.
115
+
116
+ ### `metadata()` and `proteomes()`
117
+
118
+ Both are thin passthroughs. They need no index, and they return what EBI sends.
119
+
120
+ `metadata("P69905")` calls `https://alphafold.ebi.ac.uk/api/prediction/P69905` and returns a list of records with keys such as `entryId`, `uniprotAccession`, `gene`, `taxId`, `organismScientificName`, `sequence`, `globalMetricValue` (mean pLDDT) and `latestVersion`. The API answers a well-formed accession that has no prediction with HTTP 404 and an empty JSON object; `metadata()` returns `[]` for that case. A value that is not made of letters, digits, `-` and `_` raises `InvalidIdError` (a `ValueError`) before any request is sent. An identifier the server rejects (HTTP 400) raises `ApiError` with the server's message.
121
+
122
+ `proteomes()` returns one dict per bulk tar with EBI's fields (`archive_name`, `type`, `num_predicted_structures`, `size_bytes`, and for proteome tars `species`, `common_name` and `reference_proteome`) plus `url`, the location of the tar, and `tax_id`, parsed from names such as `UP000000625_83333_ECOLI_v6.tar`. `tax_id` and `reference_proteome` are `None` for the two Swiss-Prot rows.
123
+
124
+ Both functions retry timeouts and 5xx responses twice with backoff, and never retry a 4xx.
125
+
126
+ ## How it works
127
+
128
+ 1. **The index.** One SQLite file, `index-v6.sqlite`, has a row per structure file: entry id, format, tar, byte offset and length of the gzip member. It also records each tar's size, `Last-Modified` and `ETag` at build time.
129
+ 2. **Where the index is hosted.** It is a release asset of this repository: `https://github.com/Scigantic/scigantic-alphafold/releases/download/index-v6/index-v6.sqlite.gz`. EBI does not host or maintain it.
130
+ 3. **First use.** The first call that needs the index downloads it to the cache directory, `~/.cache/scigantic-alphafold` by default. The download goes to a temporary name and is renamed into place, so an interrupted download does not leave a partial index. The download is checked against the SHA-256 published next to it in `index-v6.json`; a mismatch is refused, which can happen for a moment while the index is being republished. Later calls open the cached file. `import scigantic_alphafold` alone does not download anything.
131
+ 4. **One range read.** `fetch` looks up the entry, sends one `GET` with a `Range` header to the tar on `ftp.ebi.ac.uk`, and gunzips the bytes that come back. A member is small: the mmCIF of a 217-residue *H. pylori* protein (`AF-G5EB01-F1`) is 47,367 bytes gzipped, read out of a 173 MB tar.
132
+ 5. **Staleness check.** If the `Last-Modified` or `ETag` of the response differs from what the index recorded, the bytes do not gunzip, or the file name recorded in the gzip header is not the entry that was asked for, the offsets can no longer be trusted and the call raises `StaleIndexError`. `af.refresh_index()` downloads the current index again.
133
+
134
+ These environment variables change where the index comes from and what it may point at:
135
+
136
+ | variable | meaning |
137
+ |---|---|
138
+ | `SCIGANTIC_ALPHAFOLD_INDEX` | a local path or a URL of an index to use instead of the default, as `.sqlite` or `.sqlite.gz` |
139
+ | `SCIGANTIC_ALPHAFOLD_CACHE` | the cache directory, instead of `~/.cache/scigantic-alphafold` |
140
+ | `SCIGANTIC_ALPHAFOLD_ALLOW_ANY_HOST` | set to `1` to accept an index whose tars are not on `https://ftp.ebi.ac.uk`, for a private mirror or for tests |
141
+
142
+ By default the library only sends range requests to `https://ftp.ebi.ac.uk`, whatever the index says, so a wrong or tampered index cannot point it at another host.
143
+
144
+ `af.index_info()` reports which index is in use. The index is built by reading every tar once and recording where each member starts; [builders/README.md](https://github.com/Scigantic/scigantic-alphafold/blob/main/builders/README.md) describes the build and how to reproduce it.
145
+
146
+ ## Being a polite client of EBI
147
+
148
+ The FTP area is a shared public service that EMBL-EBI runs at its own cost. This library is meant for reading some structures, not for copying the database.
149
+
150
+ - Every request carries an honest `User-Agent`: `scigantic-alphafold/<version> (+https://github.com/Scigantic/scigantic-alphafold)`. Do not change it to look like a browser.
151
+ - `fetch_many` uses 4 connections by default. Keep `workers` near that, and do not run several processes in parallel against the FTP.
152
+ - Requests are retried with backoff only after timeouts and 5xx responses.
153
+ - If you need a whole proteome or all of Swiss-Prot, do not loop over `fetch`. Download EBI's tar once: `af.proteomes()` lists the URLs. For large transfers EBI recommends [Globus](https://app.globus.org/file-manager?origin_id=47772002-3e5b-4fd3-b97c-18cee38d6df2&origin_path=%2Fpub%2Fdatabases%2Falphafold%2F) (endpoint "EMBL-EBI Public Data", path `/pub/databases/alphafold/`), as described in the [FTP README](https://ftp.ebi.ac.uk/pub/databases/alphafold/README.txt).
154
+ - If EBI limits or closes range access to the tars, this library will stop working, and it will not try to get around that.
155
+
156
+ ## Command line
157
+
158
+ `scigantic-alphafold` wraps the same functions.
159
+
160
+ ```console
161
+ $ scigantic-alphafold get P69905 # writes the mmCIF file to the current directory
162
+ $ scigantic-alphafold get P69905 P68871 -f pdb -o structures
163
+ $ scigantic-alphafold get P69905 --stdout | head
164
+ $ scigantic-alphafold locate P69905 # tar, offset and length
165
+ $ scigantic-alphafold info # the index in use
166
+ $ scigantic-alphafold refresh # download the index again
167
+ ```
168
+
169
+ Two more subcommands, `build-part <tar>` and `merge <parts dir> -o index-v6.sqlite`, build the index; `merge --only <tar>...` builds a partial index from the tars you streamed. They are documented in [builders/README.md](https://github.com/Scigantic/scigantic-alphafold/blob/main/builders/README.md).
170
+
171
+ ## Data license and citation
172
+
173
+ This project is not affiliated with, endorsed by or supported by EMBL-EBI or Google DeepMind. It redistributes no AlphaFold DB data: the structures are read from EBI's servers at the time of the call, and the index holds only file names, byte offsets and lengths.
174
+
175
+ The [AlphaFold DB site](https://alphafold.ebi.ac.uk/) states: "Data is available for academic and commercial use, under a [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) licence." It also states: "AlphaFold Data Copyright (2022) DeepMind Technologies Limited." (both read on 2026-11-20). CC BY 4.0 requires attribution. The site's disclaimer says the data "is for theoretical modelling only" and has "not been validated for, and are not approved for, any clinical use"; read it before relying on a structure. Use of AlphaFold DB is also subject to the [EMBL-EBI Terms of Use](https://www.ebi.ac.uk/about/terms-of-use).
176
+
177
+ If you use structures obtained through this library, cite AlphaFold DB and AlphaFold, not this package. The papers the site asked users to cite on 2026-11-20:
178
+
179
+ - Bertoni D. et al. *AlphaFold Protein Structure Database 2025: a redesigned interface and updated structural coverage*. Nucleic Acids Research (2025). [academic.oup.com/nar/article/54/D1/D358/8340156](https://academic.oup.com/nar/article/54/D1/D358/8340156). Listed on the home page for use of the resource.
180
+ - Fleming J. et al. *AlphaFold Protein Structure Database and 3D-Beacons: New Data and Capabilities*. Journal of Molecular Biology (2025). [sciencedirect.com/science/article/pii/S0022283625000336](https://www.sciencedirect.com/science/article/pii/S0022283625000336). Listed on the download page for use of a prediction.
181
+ - Jumper J. et al. *Highly accurate protein structure prediction with AlphaFold*. Nature (2021). [doi:10.1038/s41586-021-03819-2](https://doi.org/10.1038/s41586-021-03819-2). Listed on the download page for predictions from the UniProt dataset.
182
+
183
+ The site's guidance can change. Check [alphafold.ebi.ac.uk](https://alphafold.ebi.ac.uk/) for the current text before you publish.
184
+
185
+ This package's code is MIT-0. That does not extend to the data.
186
+
187
+ ## Testing
188
+
189
+ ```console
190
+ $ pip install -e ".[dev]"
191
+ $ pytest
192
+ $ mypy --strict src/scigantic_alphafold
193
+ ```
194
+
195
+ The default run is offline: tests use local files and an HTTP server on localhost. Tests marked `live` send a few small requests to EBI and are skipped unless `SCIGANTIC_ALPHAFOLD_LIVE=1` is set.
196
+
197
+ ## License
198
+
199
+ MIT-0 for the code. See [Data license and citation](#data-license-and-citation) for the data.
@@ -0,0 +1,59 @@
1
+ [build-system]
2
+ # 77.0.3 is the first setuptools with SPDX license expressions and license-files (PEP 639).
3
+ requires = ["setuptools>=77.0.3"]
4
+ build-backend = "setuptools.build_meta"
5
+
6
+ [project]
7
+ name = "scigantic-alphafold"
8
+ dynamic = ["version"]
9
+ description = "Read single AlphaFold DB v6 structures with one HTTP range request against EBI's public bulk tars, using a prebuilt byte-offset index. No mirror, no account, no key, no runtime dependencies."
10
+ readme = "README.md"
11
+ requires-python = ">=3.10"
12
+ license = "MIT-0"
13
+ license-files = ["LICENSE"]
14
+ authors = [{ name = "Scigantic" }]
15
+ keywords = ["alphafold", "afdb", "protein-structure", "mmcif", "pdb", "uniprot", "structural-biology", "range-request"]
16
+ classifiers = [
17
+ "Programming Language :: Python :: 3",
18
+ "Programming Language :: Python :: 3.10",
19
+ "Programming Language :: Python :: 3.11",
20
+ "Programming Language :: Python :: 3.12",
21
+ "Programming Language :: Python :: 3.13",
22
+ "Programming Language :: Python :: 3.14",
23
+ "Typing :: Typed",
24
+ "Topic :: Scientific/Engineering :: Bio-Informatics",
25
+ "Topic :: Scientific/Engineering :: Chemistry",
26
+ ]
27
+
28
+ # Standard library only at runtime.
29
+ dependencies = []
30
+
31
+ [project.optional-dependencies]
32
+ dev = ["pytest>=7", "mypy>=1.10"]
33
+
34
+ [project.urls]
35
+ Homepage = "https://scigantic.com"
36
+ Repository = "https://github.com/Scigantic/scigantic-alphafold"
37
+ Issues = "https://github.com/Scigantic/scigantic-alphafold/issues"
38
+
39
+ [project.scripts]
40
+ scigantic-alphafold = "scigantic_alphafold.cli:main"
41
+
42
+ [tool.setuptools.packages.find]
43
+ where = ["src"]
44
+
45
+ [tool.setuptools.dynamic]
46
+ # The one place the version is written: src/scigantic_alphafold/_version.py.
47
+ version = { attr = "scigantic_alphafold._version.__version__" }
48
+
49
+ [tool.setuptools.package-data]
50
+ scigantic_alphafold = ["py.typed"]
51
+
52
+ [tool.pytest.ini_options]
53
+ testpaths = ["tests"]
54
+ markers = [
55
+ "live: talks to the real EBI server or GitHub; skipped unless SCIGANTIC_ALPHAFOLD_LIVE=1",
56
+ ]
57
+
58
+ [tool.mypy]
59
+ strict = true
@@ -0,0 +1,4 @@
1
+ [egg_info]
2
+ tag_build =
3
+ tag_date = 0
4
+
@@ -0,0 +1,57 @@
1
+ """Read single AlphaFold DB v6 structures with one HTTP range request.
2
+
3
+ import scigantic_alphafold as af
4
+
5
+ text = af.fetch("P69905") # mmCIF text
6
+ path = af.download("P69905", "structures", fmt="pdb")
7
+
8
+ The bytes come straight from EBI's public bulk tars; a prebuilt offset index says where
9
+ each structure sits. Importing this package touches neither the network nor the disk:
10
+ the index is located and opened on the first call that needs it.
11
+ """
12
+
13
+ from __future__ import annotations
14
+
15
+ from ._api import ApiError, metadata, proteomes
16
+ from ._errors import (
17
+ AlphaFoldError,
18
+ AmbiguousEntryError,
19
+ FetchError,
20
+ FetchManyError,
21
+ IndexSchemaError,
22
+ IndexUnavailableError,
23
+ InvalidIdError,
24
+ NotInBulkSetError,
25
+ StaleIndexError,
26
+ )
27
+ from ._fetch import download, fetch, fetch_bytes, fetch_many
28
+ from ._index import Member, Tar, entries, has, index_info, locate, refresh_index, tars
29
+ from ._version import __version__
30
+
31
+ __all__ = [
32
+ "AlphaFoldError",
33
+ "AmbiguousEntryError",
34
+ "ApiError",
35
+ "FetchError",
36
+ "FetchManyError",
37
+ "IndexSchemaError",
38
+ "IndexUnavailableError",
39
+ "InvalidIdError",
40
+ "Member",
41
+ "NotInBulkSetError",
42
+ "StaleIndexError",
43
+ "Tar",
44
+ "__version__",
45
+ "download",
46
+ "entries",
47
+ "fetch",
48
+ "fetch_bytes",
49
+ "fetch_many",
50
+ "has",
51
+ "index_info",
52
+ "locate",
53
+ "metadata",
54
+ "proteomes",
55
+ "refresh_index",
56
+ "tars",
57
+ ]