merck

Merck Molecular Activity Challenge code

re-implementation of the paper (the recommended model):

Ma, J., Sheridan, R.P., Liaw, A., Dahl, G.E. and Svetnik, V., 2015. Deep neural nets as a method for quantitative structure–activity relationships. Journal of chemical information and modeling, 55(2), pp.263-274.

Installation

The code was tested in Keras with Tensorflow backend. The packages needed are listed in the requirements.txt

Installing python virtual environment and requirements

pip install virtualenv
virtualenv --no-site-packages vkeras
source vkeras/bin/activate
pip install -r path/to/requirements.txt

Running the Code

• Download the training and test data-set from: Paper supplementary materials

• Set data_root and save_root variables in data_preprocessing.py and run it (This will remove the features that are not common to both training and test sets and, rescale features and activations).

  • Currently the features are rescaled to 0-1 by dividing each column by its max and the activations are rescaled to their z-score

• point the data_root in main.py to where the pre-processed training and test files are located.

• python main.py

Results

The Standard Error of Prediction (SEP) on the test set

Dataset	merk paper	This implementation
3A4	0.48	0.50
CB1	1.25	1.21
DPP4	1.30	1.68
HIVINT	0.44	0.47
HIVPROT	1.66	1.60
LOGD	0.51	0.51
METAB	21.78	23.19
NK1	0.76	0.76
OX1	0.73	0.81
OX2	0.95	0.93
PGP	0.36	0.38
PPB	0.56	0.57
RAT_F	0.54	0.55
TDI	0.40	0.41
THROMBIN	2.04	2.10