- Retrain internal model using new sequence embeddings and remove broken/duplicate BGCs from MIBiG 2.0.
- Bump minimum `pyhmmer` version to `v0.4.0` to improve exception handling.
- Bump minimum `pyrodigal` version to `v0.5.0` to fix sequence decoding on some platforms.
- Use p-values instead of e-values to filter domains obtained with HMMER.
-`gecco cv` and `gecco train` now seed the RNG with a user-defined seed before shuffling rows of training data.
### Fixed
- Extraction of BGC compositions for the type predictor while training.
-`ClusterCRF.trained` failing to open an external model.
### Added
-`Domain.pvalue` attribute to access the p-value of a domain annotation.
- Mandatory `pvalue` column to `FeatureTable` objects.
- Support for loading several feature tables in `gecco train` and `gecco cv`.
- Warnings to `ClusterCRF.fit` when selecting uninformative features.
-`--correction` flag to `gecco train` and `gecco cv`, allowing to give a multiple testing correction method when computing p-values with the Fisher Exact Tests.
### Removed
- Outdated `gecco embed` command.
- Unused `--truncate` flag from the `gecco train` CLI.
- Tigrfam domains, which is not improving performance on the new training data.
