src

This folder contains source code used by the repository.

R/python scripts

• AIDTox model

  • dtox.py learns and evaluates AIDTox model
    • dtox_data.py contains data-formatting functions used for AIDTox model training.
    • dtox_hierarchy.py contains functions used to process sorted AIDTox hiearchy files and compute model statistics.
    • dtox_nn.py contains functions used to build basic neural network structure for AIDTox model.
    • dtox_loss.py contains the the loss function used in AIDTox model.
    • early_stop.py contains early stop function of AIDTox model.
    • dtox_learning.py contains deep learning functions used in the AIDTox model construction.
    • run/run_aidtox_implementation.R generates shell scripts that run AIDTox model (DTox with chemical-gene connections from ComptoxAI) on Tox21 datasets. run/run_dtox_comparison_implementation.R generates shell scripts that run DTox on Tox21 datasets, which matches the configuration of optimal AIDTox model, for performance comparison. run/run_aidtox_null_implementation.R generates shell scripts that run AIDTox model on outcome-shuffled Tox21 datasets under Reactome pathway hierarchy.
  • predict_dtox.py implements trained AIDTox model to predict outcome probability based on input feature data.
  • interpret_dtox.py implements layer-wise relevance propagation to evaluate relevance of AIDTox paths.
    • dtox_lrp.py contains functions used for implementing LRP to evaluate relevance of AIDTox paths.
  • run/run_interpret_aidtox.R generates shell scripts that runs interpretation procedure on optimal AIDTox models trained for Tox21 datasets.

• Simple machine learning model

  • simple/simple.py develops and evaluates simple machine learning model (random forest or gradient boosting).
    • simple/simple_learning.py contains functions for building, evaluating, and implementing simple machine learning models.
    • run/run_simple.R generates shell scripts that run simple machine learning models on Tox21 datasets under different hyperparameter settings.

• Model performance analysis

  • analysis_dtox/collect_model_results.R collects machine learning model basic info and performance metrics from performance files.
  • analysis_dtox/analyze_dtox_results.py identifies optimal hyperparameter setting of machine learning method implementation, then compares and visualizes model performance across different method implementations.
    • analysis_dtox/dtox_analysis.py contains functions used in AIDTox model result anaysis.
    • analysis_dtox/dtox_plot.py contains functions for visualizing AIDTox model results.

• AIDTox feature analysis, model performance comparison, and visualization

  • analysis_aidtox/compare_top_features.py visualizes the AIDTox model performance derived from different numbers of top predictive gene features, and compares the performance by connection types.
  • analysis_aidtox/compare_model_perf.py visualizes the comparison of optimal performance between AIDTox and well-established models (DTox, RF, GB) using barplot.
  • analysis_aidtox/visualize_protein_classes.py uses Radar plots to visualize the class and subclass distribution comparison between AIDTox and DTox target features.
  • analysis_aidtox/analyze_compound_path_map.R extracts info of compounds and their identified VNN paths from AIDTox interpretation results.
  • analysis_aidtox/visualize_compound_path_map.R uses Sankey diagram to visualize VNN paths identified for HEK293 cytotoxicity outcome, which connect together compounds, gene features, pathway modules and the outcome.

• functions.R contains R functions required for other scripts in the repository.

Executable shell scripts

• AIDTox model implementation

  • run/run_aidtox_implementation_compound_gene_comptoxai_tox21.sh runs run/run_aidtox_implementation.R to generate run/aidtox_compound_gene_comptoxai_tox21/dtox_select_compound_gene_comptoxai_tox21_implementation.sh. run/aidtox_compound_gene_comptoxai_tox21/dtox_select_compound_gene_comptoxai_tox21_implementation.sh implements dtox.py on compound ComptoxAI connections-Tox21 assay outcome datasets under sorted Reactome pathway hierarchy.
  • run/aidtox_compound_gene_comptoxai_tox21_null/dtox_compound_select_gene_comptoxai_binding_tox21_null.sh implements dtox.py on compound ComptoxAI connections-Tox21 assay outcome datasets under shuffled Reactome pathway hierarchy.

• AIDTox model interpretation

  • run/interpret_dtox_compound_select_gene_comptoxai_tox21_implementation.sh implements interpret_dtox.py on optimal models trained for compound ComptoxAI connections-Tox21 assay outcome datasets.

• Compared model implementation

  • run/run_dtox_comparison_implementation_compound_gene_comptoxai_tox21.sh runs run/run_dtox_comparison_implementation.R to generate run/dtox_comparison_compound_gene_comptoxai_tox21/dtox_compound_select_gene_comptoxai_target_probability_tox21_implementation.sh. run/dtox_comparison_compound_gene_comptoxai_tox21/dtox_compound_select_gene_comptoxai_target_probability_tox21_implementation.sh implements dtox.py on compound target binding-Tox21 assay outcome datasets under sorted Reactome pathway hierarchy.
  • run/run_simple_compound_structure_tox21_comptoxai.sh runs run/run_simple.R to generate run/simple_compound_structure_tox21/simple_compound_structure_tox21_select_gene_comptoxai_binding_randomforest.sh and run/simple_compound_structure_tox21/simple_compound_structure_tox21_select_gene_comptoxai_binding_xgboost.sh. run/simple_compound_structure_tox21/simple_compound_structure_tox21_select_gene_comptoxai_binding_randomforest.sh implements simple/simple.py to build random forest models on compound structure-Tox21 assay outcome datasets under different hyperparameter settings. run/simple_compound_structure_tox21/simple_compound_structure_tox21_select_gene_comptoxai_binding_xgboost.sh implements simple/simple.py to build gradient boosting models on compound structure-Tox21 assay outcome datasets under different hyperparameter settings.

• Model performance analysis, comparison, and visualization

  • Result collection
    • run/collect_model_results_aidtox_compound_gene_comptoxai_tox21.sh implements analysis_dtox/collect_model_results.R to collect results of AIDTox models built upon compound target binding-Tox21 assay outcome datasets under sorted Reactome pathway hierarchy.
    • run/collect_model_results_dtox_comparison_compound_gene_comptoxai_tox21.sh implements analysis_dtox/collect_model_results.R to collect results of DTox comparison models built upon compound target binding-Tox21 assay outcome datasets under sorted Reactome pathway hierarchy.
    • run/collect_model_results_simple_compound_structure_tox21.sh implements analysis_dtox/collect_model_results.R to collect results of simple machine learning models built upon compound structure-Tox21 assay outcome datasets under different hyperparameter settings.
  • Result analysis
    • run/analyze_dtox_results_aidtox_compound_gene_comptoxai_tox21.sh implements analysis_dtox/analyze_dtox_results.py to identify optimal hyperparameter setting of AIDTox implementation on compound target binding-Tox21 assay outcome datasets under sorted Reactome pathway hierarchy.
    • run/analyze_dtox_results_simple_compound_structure_tox21.sh implements analysis_dtox/analyze_dtox_results.py to identify optimal hyperparameter setting of simple machine learning model implementation on compound structure-Tox21 assay outcome datasets.