In-Pero: Prediction of sub-peroxisomal localisation using deep learning embeddings and support vector machines.

To use the In-Pero.py script, you first need to download and install:

1) seqvec

Instructions are available here: https://github.com/Rostlab/SeqVec.

or

pip install seqvec

https://pypi.org/project/seqvec/

2) UniRep and the related weight files, in this case we used the 1900_weights.

https://github.com/churchlab/UniRep

Make sure you download the 1900_weights directory and place it together with the other files in this repository.

For example you can first install awscli:

pip install awscli

Then download the weights with

aws s3 sync --no-sign-request --quiet s3://unirep-public/1900_weights/ 1900_weights

3) A pre-computed model 'LR_model2.sav'/'SVM_model.sav' is also required (uncomment the selected model in the In-Pero.py script).

4) Additional requirements

Suggested packages:

• numpy 1.17.2
• biopython 1.77
• tensorflow 1.14
• pandas 0.25.1
• scikit-learn 0.22
• seqvec 0.4.1
• scipy 1.4.1
• overrides 3.1.0

Usage of In-Pero.py

Usage:

./In-Pero.py <filename>.fasta

Outputs:

• Log file ('<filename>_output.txt') containing the entries subdivided in matrix and membrane proteins.
• The UniRep encoding
• The seqvec encoding

This repository contains:

• In-Pero_models : directory containing all the pre-computed models

• Dataset : directory containing the training and validation fasta files

• LR_model2.sav : The LR pre-computed model

• SVM_model.sav : The SVM pre-computed model

• PP_matrix.fasta : fasta file containg the matrix proteins used for building the dataset

• PP_membrane.fasta : fasta file contain the membrane proteins used for building the dataset

• useful scripts retrived from the UniRep Github repository (see above):

  • data_utils.py
  • unirep.py
  • utils.py

The code in this repository is licensed under the terms of GPL v3 as specified by the LICENSE file.