Graph-Root

Root-Associated Proteins in Plants Prediction Model Based on Graph Convolutional Network

Introduction

In this repository, a model for prediction of root-related proteins in plant is provided. In this model, each protein is represented by features derived from its sequence, a protein network, and functional domains. The sequence features are refined from the raw features of amino acids, which are processed by graph convolutional network and multi-head attention, the network features are extracted from a protein network via Node2vec, and the domain features were derived from functional domain composition and further processed by singular value decomposition. All these features are fed into the fully connected layer to make predictions.

Requirements

To run this program, you may need:

• Python 3.6 or later
• Pytorch 1.12.1 and other related packages
• Windows 10 enviroment
• GPU (optional for cuda)

How to use

1. Set up your enviroment and download the code from github:

   git clone https://github.com/ken0414/Graph-Root.git

2. Put your data into the appropriate folder:

   [network embedding feature(node2vec)] --> ./data/n2v
   [domain feature(IPR)] --> ./data/ipr
   [protein structure feature] --> ./data/graph    
   [protein node feature(PSSM)] --> ./data/pssm

3. Activate your enviroment and run main.py:

   $ python main --mode cv --n2v example_n2v --run 10

Within this line of code, you can choose between cross-validation mode or independent test set mode by modifying the value after mode, n2v to change the file used, and run to select the number of repetitive runs.

option	value
mode	cvorout
n2v	filename of node2vec
run	int value for run times

4. Get result: For each fold in a single cv, you can get the best epoch of the train in train_result.txt. After all fold trained in a single cv, you can get the evaluation of all fold in predict_result.txt and the result of prediction in fold ./result. If you run on out mode, there will be only 1 result in train_result.txt and predict_result.txt. If the value of run is bigger than 1, former result in train_result.txt of a single cv or out will be override. If you want to save the result of this file, please modify the code on your own. Samely, if you run the main.py again, all the result will be override.