Antimicrobial peptide prediction with transformer and imbalanced learning

This is a method for integrating transformer and imbalanced multi-label learning to identify the functional activities of targets for antimicrobial peptides

Requirements

We have already integrate the environment in env.yaml. execute conda create -f env.yaml to install packages required in a new created DEEPSEQENV conda env.

The environment is based on pytorch=1.4.0=py3.6_cuda10.1.243_cudnn7.6.3_0. You may refer to here to custom the pytorch environment for your own machine.

Enter the enviornment with conda activate DEEPSEQENV before further executions.

Training establishment

Execute train.py to start a training process for establishing AMP prediction model with different missions (make sure the pretrained tape model is downloaded before training). For example, to train a model for AMP identification with proper hyper-parameters, you can run:

python train.py  --cuda --seed 810 --task "AMP" --lr 0.03 --ckpt-iter 60 -e 256 -b 64 -d "trial-amp"

to train a model for functional activity prediction, for instance, simply run:

python train.py  --cuda --task "mtl" -d "trial-mtl"

For more information about parser arguments, please refer to the train.py.

Evaluation

To evaluate the model, you can use evaluate.py with parsing the well-trained result path. For example, for the mentioned AMP identification trial:

python evaluate.py --path "./trial-amp" --cuda True