trialblazer 0.1.0

A Chemistry-Focused Predictor of Toxicity Risks in Late-Stage Drug Development

Badges

(Customize these badges with your own links, and check https://shields.io/ or https://badgen.net/ to see which other badges are available.)

fair-software.eu recommendations
(1/5) code repository
(2/5) license
(3/5) community registry
(4/5) citation
(5/5) checklist
howfairis
Other best practices
Documentation
Build

Data

You can download the data, including training_and_test_data, precalculated_data_for_trialblazer_model and precomputed_data_for_reproduction_with_notebooks, from: https://doi.org/10.5281/zenodo.15783346

To download the data automatically, see below the description of the Command Line Interface.

Reproduce experiments

To reproduce the experiments in the paper, you can check the notebooks here: https://github.com/molinfo-vienna/trialblazer_notebooks

How to use Trialblazer

A Chemistry-Focused Predictor of Toxicity Risks in Late-Stage Drug Development

Via Command Line

Several commands are made available:

Downloading the model

# Default model and default folder ($HOME/.trialblazer/models/base_model)
trialblazer-download

# Use other URL/folder
trialblazer-download --url=<MODEL-URL> --model-folder=<FOLDER>

Running the algorithm

The input data should be a CSV file with headers and a column named "SMILES". If present, the column "chembl_id" will also be used for the output.

The command trialblazer --help outputs:

Options:
  --input_file TEXT    Input File  [required]
  --output_file TEXT   Output File
  --model_folder TEXT  Model Folder
  --help               Show this message and exit.

The default output file is names trialblazer.csv.

As a Python library

The library containers 2 main classes:

Trialblazer

This class loads and runs the model.

from trialblazer import Trialblazer

tb = Trialblazer(input_file=<INPUT_FILE>)
tb.run()  # Includes loading of the model, creation of the classifier, and running the algorithm

df = tb.get_dataframe() # This dataframe is augmented with RDKit Mol objects, and displaying it shows the visual representation of each molecule.

tb.write(output_file=<OUTPUT_FILE>)

Trialtrainer

This class is meant to preprocess training data to recreate a model from a single CSV file (training_target_features.csv). It downloads the Chembl database, extracts relevant info, preprocesses data for active and inactive targets, and creates fingerprints files for the 3 sets of molecules (training, active, inactive).

Simply put your training_target_features.csv in your MODEL_FOLDER and run:

from trialblazer import Trialtrainer

tt = Trialtrainer(model_folder=<MODEL_FOLDER>)
tt.build_model_data()

Then you can run the algorithm using:

from trialblazer import Trialblazer

tb = Trialblazer(input_file=<INPUT_FILE>, model_folder=<MODEL_FOLDER>)
tb.run()  # Includes loading of the model, creation of the classifier, and running the algorithm

Installation

To install via PyPI, simply run:

pip install trialblazer

To install trialblazer from GitHub repository through SSH, do:

git clone [e-mail]:molinfo-vienna/trialblazer.git
cd trialblazer
python -m pip install .

or through HTTPS:

git clone https://github.com/molinfo-vienna/trialblazer_notebooks.git
cd trialblazer
python -m pip install .

Credits

This package was created with Copier and the NLeSC/python-template.

Citation

Zhang, H., Welsch, M., Schueller, W., & Kirchmair, J. (2025). Trialblazer: A Chemistry-Focused Predictor of Toxicity Risks in Late-Stage Drug Development [Data set]. Zenodo. https://doi.org/10.5281/zenodo.15783346