INSTRUCTIONS.txt

STEP 1 - Preprocess:

Inside the rdkit_preprocessing subdirectory are two main files that should be used: ecfpFingerprintGenThreaded.py and genConvMolFeatures.py.

First you will need to load Rdkit. When installed, use: source activate my-rdkit-env

ecfpFingerprintGenThreaded.py should be used to generate ECFP fingerprints before the machine learning process and saved approprately.

genConvMolFeatures.py will save a pickeled version of the data structure holding the molecule and graph structure of the molecule.

After generating the features and saving the file, you can get rid of rdkit: deactivate my-rdkit-env

STEP 2 - Deep Learnify:

The general structure of these algorithms were built around Lasagne, and modularity of layers and features was desired.

Contains custom Lasagne/Theano layers required for certain ML purposes:

lasagneCustomFingerprintLayers.py

Contains the Lasagne models for each of the deep learning modules:

lasagneModelsFingerprints.py

Wrangles data input and output to make those processes less messy:

seqHelper.py

These are then used in the actual deep learning software--the other aptly named *.py files.

STEP 3 - Interpret:

The output of the RNN and CNN files are saved for later preprocessing. Use highlight_molecule_cnn.py for the CNN predictions and highlight_smiles_rnn.py for the RNN predictions.