******************************************************************************

Author : Srivamshi Pittala (srivamshi.pittala.gr@dartmouth.edu)

Advisor : Prof. Chris Bailey-Kellogg (cbk@cs.dartmouth.edu)

Project : PECAN: Paratope and Epitope Prediction with graph Convolution Attention Network

Description : Help file for running the scripts to learn and evaluate

graph convolution networks for epitope and paratope prediction

Cite : doi: https://doi.org/10.1101/658054

******************************************************************************

This code was developed using the code provided for Fout et al., Protein Interface Prediction using Graph Convolutional Networks published in NeurIPS, 2017.

Copyright (C) <2019> <Srivamshi Pittala>

This program is free software: you can redistribute it and/or modify

it under the terms of the GNU General Public License as published by

the Free Software Foundation, either version 3 of the License, or

(at your option) any later version.

This program is distributed in the hope that it will be useful,

but WITHOUT ANY WARRANTY; without even the implied warranty of

MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the

GNU General Public License for more details.

You should have received a copy of the GNU General Public License

along with this program. If not, see https://www.gnu.org/licenses/.

#*********************** Input Files (pickle format). Can be downloaded from zenodo here: https://zenodo.org/record/3885236.

1. Paratope prediction: results_create_files_for_paratope/
2. Epitope prediction: results_create_files_for_epitope/

The data format is the following:

• list of dictionaries, each corresponding to a protein
• • Each dictionary has the following entries(keys):
• • • complex_code: str
• • • l_vertex: numpy matrix of residues in primary protein (dimensions: num_residues x 63)
• • • • Indices [0,20] represent one hot encoding of the residue. Last entry for unresolved amino acid '*'.
• • • • Indices [21,40] represent PSSM entry for the
• • • • Indices [41:42] represent solvent accessibility
• • • • Indices [43:62] represent neighborhood composition
• • • l_edge: numpy matrix of edges, currently empty as we don't consider edge features (dimensions: num_residues x 25 x 2)
• • • l_hood_indices: numpy matrix specifying spatial neighborhood for graph convolution (dimensions: num_residues x 25)
• • • label: numpy matrix with labels {1: interface, -1: non-interface} (dimensions: num_residues x 2)
• • • r_vertex: same as l_vertex, but for secondary protein
• • • r_edge: same as l_edge, but for secondary protein
• • • r_hood_indices: same as l_hood_indices, but for secondary protein
• • • label_r: same as label, but for secondary protein #***********************

#----------------------- Software (version) #-----------------------

1. Python 2.7.15
2. PyYAML 3.13
3. Numpy 1.14.5
4. Pandas 0.23.4
5. Scikit-learn 0.19.1
6. tensorflow 1.10.0 (gpu mode)
7. cudatoolkit 8.0.61
8. cudnn 5

#*********************** Training base network on proteins #***********************

1. cd GCN_protein_base
2. Train "No convolution" network: python sample_experiment_base.py
3. Train "Convolution 1-layer" network: python sample_experiment_conv1.py
4. Train "Convolution 1-layer + Attention" network: python sample_experiment_attn1.py
5. Train "Convolution 2-layer" network: python sample_experiment_conv2.py
6. Train "Convolution 2-layer + Attention" network: python sample_experiment_attn2.py

#*********************** For Epitope prediction #***********************

#******** Task-specific learning #********

1. cd GCN_task
2. Change 'mode_experiment, 'data_directory' and 'output_directory' in configuration.py for epitope data
3. Train "No convolution" network: python sample_experiment_base.py
4. Train "Convolution 1-layer" network: python sample_experiment_conv1.py
5. Train "Convolution 1-layer + Attention" network: python sample_experiment_attn1.py
6. Train "Convolution 2-layer" network: python sample_experiment_conv2.py
7. Train "Convolution 2-layer + Attention" network: python sample_experiment_attn2.py

#******** Transfer learning (Make sure the base network on proteins was trained successfully) #********

1. cd GCN_xTransfer
2. Change 'mode_experiment, 'data_directory' and 'output_directory' in configuration.py for epitope data
3. Train "No convolution" network: python sample_experiment_base.py
4. Train "Convolution 1-layer" network: python sample_experiment_conv1.py
5. Train "Convolution 1-layer + Attention" network: python sample_experiment_attn1.py
6. Train "Convolution 2-layer" network: python sample_experiment_conv2.py
7. Train "Convolution 2-layer + Attention" network: python sample_experiment_attn2.py

#*********************** For Paratope prediction #***********************

#******** Task-specific learning #********

1. cd GCN_task
2. Change 'mode_experiment, 'data_directory' and 'output_directory' in configuration.py for paratope data
3. Train "No convolution" network: python sample_experiment_base.py
4. Train "Convolution 1-layer" network: python sample_experiment_conv1.py
5. Train "Convolution 1-layer + Attention" network: python sample_experiment_attn1.py
6. Train "Convolution 2-layer" network: python sample_experiment_conv2.py
7. Train "Convolution 2-layer + Attention" network: python sample_experiment_attn2.py

#******** Transfer learning (Make sure the base network on proteins was trained successfully) #********

1. cd GCN_xTransfer
2. Change 'mode_experiment, 'data_directory' and 'output_directory' in configuration.py for paratope data
3. Train "No convolution" network: python sample_experiment_base.py
4. Train "Convolution 1-layer" network: python sample_experiment_conv1.py
5. Train "Convolution 1-layer + Attention" network: python sample_experiment_attn1.py
6. Train "Convolution 2-layer" network: python sample_experiment_conv2.py
7. Train "Convolution 2-layer + Attention" network: python sample_experiment_attn2.py