Readme

Requirements

1. numpy==1.19.4
2. pandas==0.25.3
3. python==3.6.12
4. scikit-learn==0.23.2
5. torch==1.7.0+cu101
6. torchvision==0.8.0+cu101
7. xgboost==1.3.0
8. tqdm==4.54.1

Installation

pip install -r requirements.txt

Data preparation

Generate training and testing files

python generate_sample.py

Generate sequences for searching

Use sequence_generated.py in ./sequence_generated to generate the sequence for customized searching space, we offered sequences for peptides which length is 6 and the script to generate peptide sequences of length 7 in folder ./sequence_generated.

Generate strutual data for sequences

Use cal_pep_des.py in ./featured_data_generated to generate structual data for Classification and Ranking stage from the sequences derived in the last step.

Model Training

Pipeline training

Use train.py to get all the params for the three models(Classifcation, Ranking, Regressing). You can use customized training data or data generated from Grampa dataset.

Incremental learning

Use lstm_fine_tune.py for incremental learning. The augmented data was provided in folder ./data/origin_data. Using customized data validated in other wet-lab settings is optional.

Searching for antimicrobial sequences

Use predict.py to get the final searching result. For a vast searching space, you may use 'chunk' mechanism to avoid RAM shortage.