Changes in version 0.3.0                        

Data

All tables rebuilt from Ensembl 116 (previously Ensembl 114). Recipes
are pinned to the June 2026 archive, jun2026.archive.ensembl.org.

Recipes select data by species, and BioMart always returns whatever
assembly is current for that species, so several datasets moved to a new
assembly with this release:

| Dataset | Species | Assembly now served | |---|---|---| | grch38 |
Human | GRCh38 | | grch37 | Human | GRCh37 | | grcm39 | Mouse | GRCm39 |
| grcr8 | Rat | GRCr8 | | grcg7b | Chicken | bGalGal1.mat.broiler.GRCg7b
| | wbcel235 | Roundworm | WBcel235 | | bdgp654 | Fruitfly | BDGP6.54 |
| grcz11 | Zebrafish | GRCz11 | | mmul10 | Macaque | Mmul_10 | |
roscfam1 | Dog | ROS_Cfam_1.0 | | sscrofa111 | Pig | Sscrofa11.1 |

Breaking: datasets renamed after their assemblies

Datasets are named after the genome assembly from now on, not the
species.

A species-based name stops describing its own contents as soon as
Ensembl switches assembly, and nothing fails to make that visible — the
data simply changes underneath the name. Naming after the assembly means
a stale name is obvious.

Eight datasets were renamed. The old names were removed rather than
deprecated, so existing code errors instead of quietly returning a
different assembly:

| Old | New | Species | Assembly | |---|---|---|---| | rnor6 | grcr8 |
Rat | GRCr8 | | grcm38 | grcm39 | Mouse | GRCm39 | | galgal5 | grcg7b |
Chicken | bGalGal1.mat.broiler.GRCg7b | | mmul801 | mmul10 | Macaque |
Mmul_10 | | cfamiliaris | roscfam1 | Dog | ROS_Cfam_1.0 | | bdgp6 |
bdgp654 | Fruitfly | BDGP6.54 | | drerio | grcz11 | Zebrafish | GRCz11 |
| sscrofa | sscrofa111 | Pig | Sscrofa11.1 |

This applies to the _tx2gene tables too. grch37, grch38 and wbcel235
were already assembly-named and are unchanged.

Build

  - Fixed HTTP 405 Method Not Allowed from getBM(). www.ensembl.org
    308-redirects BioMart to a GET-only endpoint, which rejects the POST
    biomaRt uses to submit query XML.
  - get_mart() constructs the Mart object directly and calls
    useDataset() on it, instead of useMart(). The BioMart registry
    advertises host="www.ensembl.org" for every mart, so useMart() would
    send its dataset-list request to the broken host before returning;
    patching mart@host afterwards was too late.
  - Rebuilt for resilience against dropped connections. Each table is
    downloaded, saved and logged individually rather than saving
    everything at the end, so an interrupted run keeps what finished.
    Completed tables are recorded in data-raw/build_progress.log and
    skipped on rerun; the log is removed once all tables are present.
  - Retries with backoff on 502/503/504/429, plus a 3s delay between
    requests and a per-host mart cache.
  - Retries also cover "is not valid", "unavailable" and "incompatible
    version". An overloaded archive serves an HTML error page with
    HTTP 200, which biomaRt parses as data and misreports as a bad
    dataset name or a version mismatch.
  - Set options(timeout = 600). biomaRt uses R's 60s default, and a cold
    archive host was measured taking 51.7s to return the registry alone.
  - tx2gene no longer overwrites recipes in place.
  - load_recipes globs *.yml only.
  - Added data-raw/switch-host.sh and data-raw/recipes/README.md.

                        Changes in version 0.2.0                        

  - Added host to data retrieval function
  - Updated all recipes to use https