dom-adv-sigU

Using Domain-Adversarial NN to Predict Cleavage Sites of Signal Peptides.

Setup environment

$ source config_env

Install

$ pip3 install -r requires.txt

Execute

Generate 20 models and evaluate them.

All available dataset: euk, gram-, gram+, all, bacteria.

Ratio: the ratio of good and bad features.

e.g. 0.5 for 1:1, 0.2 for 1:4.

$ ./run.sh euk 0.5

All the result will be in log/euk_0.5.txt.

Build Feature

Build initial features

Go to bin:

$ cd bin/

All available dataset: euk, gram-, gram+, all, bacteria.

$ ./build_train_features.py euk

Initial training features saves in data/features/ as train.npy.

Evalate features saves in data/features/ and eval.npy.

All available dataset: euk, gram-, gram+, all, bacteria.

$ ./build_eval_features.py euk

Testing features saves in data/features/ as test.npy.

Build 'good' and 'bad' features using initial features

Ratio : the ratio of good and bad features.

e.g. 0.5 for 1:1, 0.2 for 1:4

$ ./random_select.py 0.5

'Good' and 'bad' features saves in data/features/ train_good.npy and train_bad.npy.

Build 'good' and 'bad' features with SPDS17 features

Ratio : the ratio of good and bad features.

e.g. 0.5 for 1:1, 0.2 for 1:4

SPDS17 features: test.npy

$ ./con_bad_data.py 0.5

'Good' and 'bad' features saves in data/features/ train_good.npy and train_bad.npy.

Run Experiment

Train a model using 'Good' and 'bad' features

$ cd experiment/
$ ./domain_adversarial.py model_name/

The model saves in models/modle_name/

Evaluate the model

./get_pred.py model_name/