This repository contains a simple implementation of the naive Bayes classifier to provide baseline models for the 12 datasets in the Tox21 Challenge. The implementation makes use of LyChI for structure standardization and PCFP for descriptor/feature extraction. Please note that this code is for demonstration purposes only; we make no claims as to its validity or correctness. We do, however, encourage you to apply the models as-is (see below for instructions) and submit the results to get a feel for what to expect.
The code is self-contained in that all you need is a recent version of
the Java development kit (e.g.,
JDK 6+) and Apache ant. To compile the code,
simply type ant at the command line. This step is optional as
the code is automatically compiled when any of the commands below is
invoked.
The code comes bundled with prebuilt models under the models
directory. However, if you want to build new set of models based on
tweaks you've made to the code, simply type the following on the
command line:
ant train
If all goes well, this command will generate the following 12 files, one for each respective model.
NR-AHR.props
NR-AR-LBD.props
NR-AR.props
NR-AROMATASE.props
NR-ER-LBD.props
NR-ER.props
NR-PPAR-GAMMA.props
SR-ARE.props
SR-ATAD5.props
SR-HSE.props
SR-MMP.props
SR-P53.props
To test these models, simply move them to the models directory
(thereby replacing the old ones) and type ant predict (see
below) to test out the new models.
To apply the models to test dataset, simply run the following command:
ant predict
Upon successful execution, the following files are generated:
NR-AHR.txt
NR-AR-LBD.txt
NR-AR.txt
NR-AROMATASE.txt
NR-ER-LBD.txt
NR-ER.txt
NR-PPAR-GAMMA.txt
SR-ARE.txt
SR-ATAD5.txt
SR-HSE.txt
SR-MMP.txt
SR-P53.txt
Each file is in the proper format for [submission] (https://tripod.nih.gov/tox21/challenge/submissions.jsp).
For questions and/or problems with the code, please feel free to contact the Tox21 team at ncats9800tox21challenge@mail.nih.gov. Good luck with the competition!