Author: Sean Davern
Desciption: This project aims at using dense neural networks to model chemical compound toxicities based on physiochemical and stereochemical properties. It was the fifth and final project of my Metis Data Science boot camp.
Documentation for this project is represented by Jupyter Notebook (.ipynb files) comments, the slide presentation in ./reports/slide_deck.pptx (and .pdf) and in the summary report in ./reports/summary.pages (and .pdf). Finally, see brief explanations of what project work each file is associated with below.
Data used for this project came from the Institute of Bioinformatics Johannes Kepler University which provided data from the National Institutes of Health's 2014 Tox21 Data Challenge.
Reference:
[Mayr2016] Mayr, A., Klambauer, G., Unterthiner, T., & Hochreiter, S. (2016). DeepTox: Toxicity Prediction using Deep Learning. Frontiers in Environmental Science, 3:80. doi/10.3389/fenvs.2015.00080
[Huang2016] Huang, R., Xia, M., Nguyen, D. T., Zhao, T., Sakamuru, S., Zhao, J., Shahane, S., Rossoshek, A., & Simeonov, A. (2016). Tox21Challenge to build predictive models of nuclear receptor and stress response pathways as mediated by exposure to environmental chemicals and drugs. Frontiers in Environmental Science, 3:85.
(generated with datasciencemvp)
(modified from cookiecutter-datascience)
.
├── GPU_attempt.ipynb My attempt to use my mac's GPU
├── Imbalance+metric.ipynb Working out balancing classes in data and using alternate
│ metric for Keras optimizer
├── LICENSE
├── README.md This file
├── bayesian_optimization_singe_dnn.ipynb Implementation of hyperparameter tuning with
│ HyperOpt for the NR.AhR
├── data
│ ├── interim
│ ├── processed Each subfolder named with a numeral contains pickled pandas dataframes
│ │ │ containing a single compound. The numeral is the number of targets the compound
│ │ │ test positive for
│ │ ├── 1
│ │ │ ├── NCGC00260696-01.pkl
│ │ │ ├── ...108 other similar files
│ │ │ └── NCGC00357289-01.pkl
│ │ ├── 2
│ │ │ ├── NCGC00260789-01.pkl
│ │ │ ├── ...51 other similar files
│ │ │ └── NCGC00357254-01.pkl
│ │ ├── 3
│ │ │ ├── NCGC00260731-01.pkl
│ │ │ ├── ...29 other similar files
│ │ │ └── NCGC00357288-01.pkl
│ │ ├── 4
│ │ │ ├── NCGC00260831-01.pkl
│ │ │ ├── ...17 other similar files
│ │ │ └── NCGC00357284-01.pkl
│ │ ├── 5
│ │ │ ├── NCGC00261052-01.pkl
│ │ │ ├── ...8 similar itesm
│ │ │ └── NCGC00357109-01.pkl
│ │ ├── 6
│ │ │ ├── NCGC00261776-01.pkl
│ │ │ ├── NCGC00357007-01.pkl
│ │ │ ├── NCGC00357011-01.pkl
│ │ │ └── NCGC00357249-01.pkl
│ │ ├── 7
│ │ │ ├── NCGC00261332-01.pkl
│ │ │ ├── NCGC00261662-01.pkl
│ │ │ └── NCGC00261683-01.pkl
│ │ ├── 8
│ │ │ ├── 3-Chloro-4-methyl-7-hydroxycoumarin\ |\ C10H7ClO3\ -\ PubChem.webloc
│ │ │ ├── NCGC00357111-01.pkl
│ │ │ └── SID\ 251919981\ -\ PubChem.webloc
│ │ ├── modeled_feature_names.pkl A pickled list of names of the targets
│ │ └── non-toxic
│ │ ├── NCGC00260691-01.pkl
│ │ ├── ...414 other similar files
│ │ └── NCGC00357287-01.pkl
│ └── raw
│ ├── tox21 This folder is what tox21.zip expands to (it is excluded from Github)
│ │ ├── sampleCode.R
│ │ ├── sampleCode.py
│ │ ├── tox21.sdf.gz
│ │ ├── tox21_compoundData.csv
│ │ ├── tox21_dense_test.csv.gz
│ │ ├── tox21_dense_train.csv.gz
│ │ ├── tox21_labels_test.csv.gz
│ │ ├── tox21_labels_train.csv.gz
│ │ ├── tox21_sparse_colnames.txt.gz
│ │ ├── tox21_sparse_rownames_test.txt.gz
│ │ ├── tox21_sparse_rownames_train.txt.gz
│ │ ├── tox21_sparse_test.mtx.gz
│ │ └── tox21_sparse_train.mtx.gz
│ └── tox21.zip The data as provided by
├── flask_app
│ ├── html
│ │ ├── index.html
│ │ ├── javascripts
│ │ │ └── main.js
│ │ └── stylesheets
│ │ └── main.css
│ ├── lr.pkl This is the model the original flask app example used.
│ ├── models A set of my models that predict toxicity for the indicated target.
│ │ ├── NR_AR.h5
│ │ ├── NR_AR_LBD.joblib
│ │ ├── NR_AhR.joblib
│ │ ├── NR_Aromatase.joblib
│ │ ├── NR_ER.joblib
│ │ ├── NR_ER_LBD.joblib
│ │ ├── NR_PPAR_gamma.h5
│ │ ├── SR_ARE.joblib
│ │ ├── SR_ATAD5.joblib
│ │ ├── SR_HSE.joblib
│ │ ├── SR_MMP.joblib
│ │ └── SR_p53.joblib
│ ├── predictor_api.py The primary flask app python file
│ ├── predictor_app.py The python methods that support the predictions (incomplete)
│ ├── templates
│ │ └── predictor.html
│ └── uploads A molecule parameters file uploaded using the app
│ └── NCGC00357111-01.pkl
├── individual_compound_generation.ipynb Used to generate all the /Data/Processed/ files
├── input_parameter_exploration.ipynb Used to do EDA
├── keras_save-load_issue.ipynb Documents the Keras issue with loading models with
│ and InputLayer layer
├── main.py Not used
├── matthews_correlation_coeff.ipynb Used to explore using the Matthews Corr Coef
├── model_performance_tables.ipynb Allow easy exploration of all obtained model results
├── models Contains pickled model objects for each target and summary tables
│ ├── NR_AR
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_AR.pkl
│ ├── NR_AR_LBD
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_AR_LBD.pkl
│ ├── NR_AhR
│ │ ├── DNN0.h5
│ │ ├── DNN1.h5
│ │ ├── DNN1y.joblib
│ │ ├── DNN1z.pkl
│ │ ├── DNN2.h5
│ │ ├── DNN3.h5
│ │ ├── DNN4.h5
│ │ ├── DNN5.h5
│ │ ├── DNN6.h5
│ │ ├── DNN7.h5
│ │ ├── DNN_modT0.h5
│ │ ├── DNNtest0.h5
│ │ ├── DNNtest1.h5
│ │ ├── DNNtest2.h5
│ │ ├── RF0.joblib
│ │ ├── RF0.joblib\ copy
│ │ ├── tracking-v1.csv
│ │ ├── tracking.csv
│ │ ├── trials.pkl
│ │ └── trials_v1.pkl
│ ├── NR_AhR.pkl
│ ├── NR_Aromatase
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_Aromatase.pkl
│ ├── NR_ER
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_ER.pkl
│ ├── NR_ER_LBD
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_ER_LBD.pkl
│ ├── NR_PPAR_gamma
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── NR_PPAR_gamma.pkl
│ ├── SR_ARE
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── SR_ARE.pkl
│ ├── SR_ATAD5
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── SR_ATAD5.pkl
│ ├── SR_HSE
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── SR_HSE.pkl
│ ├── SR_MMP
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── SR_MMP.pkl
│ ├── SR_p53
│ │ ├── DNN0.h5
│ │ └── RF0.joblib
│ ├── SR_p53.pkl
│ └── first_model.h5
├── mvp.ipynb Contains my MVP
├── performance_summary.ipynb Generates the summary of performance for all targets
├── predictor.ipynb Intended to develop the predictor function for the flask app
├── proposal.md The original project proposal in markdown format
├── proposal.pdf The original project proposal in pdf format
├── random_forests.ipynb Fits random forest to all targets
├── references Technical references relevant to the project
│ ├── Analytical\ Tests Details about the analytical tests behind targets
│ │ ├── Tox21\ Data.webloc
│ │ └── tox21-ahr-p1
│ │ ├── tox21-ahr-p1.aggregrated.txt
│ │ ├── tox21-ahr-p1.description.txt
│ │ ├── tox21-ahr-p1.slp.doc
│ │ └── tox21-ahr-p1.txt
│ ├── DeepTox-\ Deep\ Learning\ for\ Toxicity\ Prediction.webloc
│ ├── Extended-Connectivity\ Fingerprints\ -\ ECFPs.webloc
│ ├── Frontiers\ |\ DeepTox-\ Toxicity\ Prediction\ using\ Deep\ Learning\ |\ Environmental\ Science.webloc
│ ├── Graph_kernels_for_chemical_informatics\ -\ DFS.pdf
│ ├── Handbook\ of\ Molecular\ Descriptors\ -\ Roberto\ Todeschini,\ Viviana\ Consonni\ -\ Google\ Books.webloc
│ ├── JKU_site
│ │ ├── DeepTox_\ Deep\ Learning\ for\ Toxicity\ Prediction.pdf
│ │ ├── Huang\ etal.pdf
│ │ └── Mayr-etal.pdf
│ ├── NIH_site
│ │ ├── about.pdf
│ │ ├── data.pdf
│ │ ├── home.pdf
│ │ └── leaderboard.pdf
│ ├── Other\ Literature
│ │ ├── PR\ and\ ROC\ Curves\ -\ Davis\ and\ Goadrich.pdf
│ │ └── Receiver\ operating\ characteristic\ -\ Wikipedia.webloc
│ ├── PyBioMed\ Chem.pdf
│ ├── Wikipedia\ -\ Drug\ Discovery\ Datasets.webloc
│ ├── data_dictionary
│ ├── github-gadsbyfly-PyBioMed.webloc
│ └── purchased_articles
│ └── Kola_et_al-2004-Nature_Reviews_Drug_Discovery.pdf
├── reports Documentation generated for this project
│ ├── figures
│ │ ├── Confusion\ Matrix.xlsx
│ │ ├── NR.AhR_DNN_PrecisionRecallCurve2.svg
│ │ ├── NR.AhR_RandomForest_PrecisionRecallCurve\ copy.svg
│ │ ├── NR.AhR_RandomForest_PrecisionRecallCurve.svg
│ │ └── overall_performance.svg
│ ├── slide\ deck.pptx The final presentation slide deck
│ ├── summary.pages The project final summary in mac Pages format
│ ├── summary.pdf The project final summary in pdf format
│ ├── templates Slide templates used or considered
│ │ ├── Microbiology\ Breakthrough\ by\ Slidesgo.pptx
│ │ └── TS101967975.potx
│ └── tree.txt The command line generated version of this document tree.
├── sample.ipynb The sample notebook provided by Johannes Kepler University
├── sample_NR.AhR_modified_metric.ipynb Generates Precision-Recall Curves for the
│ the Random Forest Models as developed from sample.ipynb
├── single_task_dnn_base.ipynb The extension of MVP.ipynb to all targets
└── src Python source code
├── __init__.py
├── explore.py
├── helper_functions.py Functions for loading data files, loading/saving
│ models and model performance dataframes
├── interpret.py
├── metrics.py Functions for calculating DNN training metrics
├── model.py
├── obtain.py
├── scrub.py
└── utils
├── __init__.py
├── load_or_make.py
└── make_dataset.py