459below/Project-CS230

Project-CS230

Project-CS230

Title: Using Deep Learning to Predict Toxicity and Lipophilicity from Molecular Fingerprints and 2D Structures.

Please download the sdf datafiles from the following links and add them to /data folder:

• tox21db: https://tripod.nih.gov/tox21/challenge/download?id=tox21_10k_data_allsdf&sec=

• ncidb: https://cactus.nci.nih.gov/download/nci/ncidb.sdf.gz

Python version supported: Python 3.

Setup Steps:

1. Install miniconda (please see instructions in https://docs.conda.io/en/latest/miniconda.html )
2. Create a new environment: conda create -n myenv python
3. Activate conda environment: source activate myenv
4. Install RDKit (necessary for sdf files preprocessing): conda install -c rdkit rdkit
5. Install requirements: pip install -r requirements.txt

Using the models

The best performing models are the following:

• Lipophilicity predictors:

  • fingerprints input
    fc_nn_6l_logp.py (LogP)
    fc_nn_6l_exp_logp.py (experimental LogP)
  • 2D molecular images input
    incep_resnet_compact_v4_logp.py (LogP)
    incep_resnet_compact_v4_exp_logp.py (experimental LogP)

• Toxicity predictors:

  • fingerprints input
    fc_nn_tox21.py

  • 2D molecular images input
    incep_resnet_tox21_t.py

General instructions set Train=True in main function for training and Train=False for running in test mode specifying a previously saved weight file. N.B: We include some of the best performing models weights files in weights/ folder.

You can see come of the previous run results (metrics images and history files) in the output/ folder.