Skip to content

Scripts

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

Scripts

README.md

Folder description

Code used to generate the results described in the study. Files containing "pipeline" in their name are meant to be used as command line tools.

Files documentation

  • analyze_hts_pipeline.py: Automates the use of MVS-A on new datasets. See the Tutorial section on the README in the main page or the documentation in the file for information on how to use it.

  • cleanup_pipeline.py: Executes the standardization and assay merging script to generate datasets ready to be analyzed. Outputs are saved in Datasets, logs are saved in Logs/cleanup.

  • info_pipeline.py: Collects relevant statistics from all datasets in Datasets. Outputs are saved in Logs

  • eval_pipeline.py: Collects performance metrics for AIC retrieval algorithms for the benchmarks in Datasets. Performance is evaluated in terms of Precision (K=10%), Enrichment Factor (K=10%), BEDROC (alpha=20), calculation time (s) and Murko scaffold diversity. Outputs are saved in Results, raw predictions are saved in /Logs/eval. Current AIC retrieval algorithms include MVS-A, CatBoost Object Importance, structural alerts (GSK and REOS filters), Isolation Forest, Variational Autoencoder and sorting by primary assay readout (top 10% most active compounds in primary screen are flagged as TPs).

  • opt_pipeline.py: Script to generate the autofluorescence predictor with Bayesian hyperparameter optimization. The 10-fold cross-validation performance is saved in Misc.

  • predictors_pipeline.py: Computes performance metrics from the predictions of Hit Dexter, SCAM Detective and the autofluorescence predictor. The predictions are stored in Logs/eval, while the output is saved in Results/fp_detectors.

  • summarize_pipeline.py: Averages performance metrics for each method across replicates and generates final summary files for performance metrics and statistical tests. The output is saved in Results/summary.

  • case_study_pipeline.py: Script to reproduce the case study analysis reported in the publication.

  • evals.py: Contains the dataset evaluation functions used to collect performance metrics in eval_pipeline.py:.

  • utils.py: Contains various helper functions used in all pipelines (i.e. generating ECFPs etc). One key function is process_ranking, which is used to convert raw importance scores into binary labels for precision@10 calculation.

  • fragment_filter.py: Contains the function for using structural alerts as decision rules to classify FPs and TPs. Currently supports GSK and REOS filters.

  • filters.cvs: SMARTS definition of the GSK structural alerts used in fragment_filter.py.

  • vae.py: PyTorch implementation of a SMILES-based Variational Autoencoder for anomaly detection. The implementation and hyperparameters are based on https://github.com/aspuru-guzik-group/chemical_vae.

  • MVS_A.py: Contains the sample_analysis class, which contains all relevant methods for running MVS-A outside of these scripts. See the README on the main page for a jupyter notebook example.

Getting help

For more technical details, check the documentation and comments provided in the respective files. For the command line tools, you can also use the --help flag to further clarify their usage. Here is an example:

python3 eval_pipeline.py --help