SigUNet

A signal peptide predictor based on deep learning.

Python version support

Python 3.6 only

Installation

For CPU: pip3 install -r requirement.cpu.txt

For GPU (suggest): pip3 install -r requirement.gpu.txt

Dataset

• SignalP dataset : SignalP 4.0 Server
• SPDS17 dataset : DeepSig dataset

Notice : Put files of SignalP dataset into data/SignalP and put files of SPDS17 dataset into data/SPDS17

Training

1. Preprocessing dataset:

./build_SignalP_dataset.py [dataset]
./build_SPDS17_dataset.py [dataset]

[dataset]: euk, gram+, gram-

2. Train models:

./nested_cv.py params/params.example.json [model_path]

For custom search space of m, n and filer size descripted in the paper, you can use your params file. Please make sure the format of your file is the same as params/params.example.json.

It will show the result of SignalP dataset after training.

3. Evaluate SPDS17 dataset

./evaluate_SPDS17.py [model_path]