Author: Wen Torng (05/2017)
Requirement: Theano (version 0.8.2), Python 2.7.15, PyTables

=========================
==== DATA PROCESSING ====
=========================

Step 1. Sample positions from each training PDB file and extract local boxes

python generate_backbone_box_20A.py dataset

dataset can be 'train' or 'test'

- Basically the same process as the full sidechain code but leave out all the sidechains. 
- Sample positions from each training PDB file and extract local boxes. Dump individual numpy arrays for every 1,000 boxes (or 100 boxes for test) per amino acid type. 

Step 2. Take in numpy arrays generated from Step 1 and generate balanced numpy arrays.

python integrate_data.py


Step 3. Write the arrays from Step 2 into pytables.

python generate_pytable_train.py
 

========================
==== MODEL TRAINING ====
========================
THEANO_FLAGS=floatX=float32 python 3DCNN.py

=========================
===== MODEL TESTING =====
=========================
THEANO_FLAGS=floatX=float32 python eval_3DCNN.py