Changes with the release of v2.0.1:

• The use of the asterisk (*) character to denote a stop codon in input peptide sequences used for training and/or evaluating is supported in this release. Users are welcomed to use them as an indications for peptides that are not truncated. However, AMPlify would remove the asterisks if they are at the end of the sequences and not consider them during deep learning processes.

• Additional scripts and data for the AMP mining workflow that has been utilized to mine novel AMPs from the UniProtKB/Swiss-Prot database have been provided in the auxiliary folder.

• Besides the original model (balanced model) in v1.0, an additional model (imbalanced model) is provided in this version.
• Add imbalanced training and test sets.
• New option -sample_ratio in the training module enables users to train their models in balanced or imbalanced settings.
• Remove option -md in the prediction module.
• Update the function of option -m in the prediction module; users can specify which model (balanced/imbalanced) to choose for prediction.
• New option -sub enables users to see predictions from sub-models.

• Add the log scaled AMPlify score to the output, calculated as -10*log10(1-Probability_score).
• Add length and charge information to the output.

• Users won't have to filter invalid sequences on their own before running the model anymore.
• Results for all invalid sequences will be filled with 'NA'.

• New option -att enables users to make their own choice whether to show attention scores.
• -of option now supports txt or tsv as the output format (the default format has been set to be tsv).
• Now users do not have to specify the model directory with -md if using the pre-trained models.
• Additional scripts and data for the AMP discovery pipeline that has been utilized to mine novel AMPs from the bullfrog genome have been provided in the auxiliary folder.

Initial release

For details, please refer to the README.md.