deepFPlearn - datasets, models, and configurations

This repository contains

• all datasets used for training and testing the models
• all trained models, and
• the JSON configuration files used to train or test the models,

as described in the original publication.

Datasets

• dataset_train_ae.[pkl,tsv]: Data used for training the generic autoencoder
• dataset_train_fnn.[pkl,csv]: Data used for training the feed forward network (classification task)
• dataset_comptox.[pkl,tsv] : Unlabeled data used to provide first predictions
• dataset_predictions_comptox_all.csv : Results of predicting the unlabeled data with all models individually

Models

• model_trained_ae*: trained Autoencoder (ae) model as published in https://doi.org/10.1093/bib/bbac257
  • model_trained_ae_weights.hdf5: the weights of the trained model, represents input for the prediction tasks
  • model_trained_ae_saved_model.tar: the full model, can also be loaded and used for prediction
• model_trained_fnn_[AR, ER, GR, TR, Aromatase, PPARg, ED]*: Classification models as published in https://doi.org/10.1093/bib/bbac257 for the different targets androgen (AR), estrogen (ER), glucocorticoid (GR), and thyroid receptors (TR), Aromatase, PPARg, and more generally with endocrine disruption (ED).
  • model_trained_fnn_[target]_model.weights.hdf5: the weights of the trained model, represents input for the prediction tasks
    
  • model_trained_fnn_[target]_saved_model.tar: the full model, can also be loaded and used for prediction
  • model_trained_fnn_[target]_history.[csv,svg]: the training history (all logged metrics) as table and x-y plot
  • model_trained_fnn_[target]_predicted.testdata: the predictions on the test data after training

Configuration files

• configFile_train_ae.json: Train the autoencoder
• configFile_train_fnn.json: Train the feed forward network without including the autoencoder (on uncompressed fingerprints)
• configFile_train_fnn_compressed.json: Train the feed forward network including the autoencoder (on compressed/encoded fingerprints)
• configFile_predict_comptox_[target].json: Use the trained autoencoder and each (w.r.t. target) of the trained feed forward models to predict the unlabeled comptox dataset