pKaPredict

📦 Overview

pKaPredict is a cheminformatics tool that predicts the acid dissociation constant (pKa) of chemical compounds from their SMILES strings. The package is designed to be easily pip-installable and modular, making it ideal for cheminformatics applications. The package combines RDKit generated molecular descriptors with a pre-trained LGBMRegressor model for fast, reproducible predictions. The workflow includes training data preprocessing, descriptor generation, model loading, prediction, and result visualization.

💡 Main Functions :

Functions used to train the machine learning model on a training dataset

→ download_raw_dataset Downloads a CSV file from a given URL and saves it in the specific data directory.

→ preview_data Loads and previews the raw pKa dataset from a relative file path.

→ clean_and_visualize_pka Cleans the training dataset by removing duplicates and missing values; also provides basic visualizations for data exploration.

→ RDKit_descriptors Transforms a DataFrame of SMILES into molecular descriptors, used for training the model.

→ plot_data Creates scatter plots comparing predicted vs. experimental pKa values to evaluate model performance and choose the most adapted model for this usage.

→ plot_k_vs_r2_ET and LGBMplot_k_vs_r2 Plots R² scores versus number of selected features (k) and saves the plot for the two machine learning models tested.

Core functions for predicting pKa from user-provided SMILES inputs

→ load_model Loads the pre-trained LightGBM model optimized for pKa prediction.

→ smiles_to_rdkit_descriptors Converts a user-provided SMILES input into a vector of RDKit molecular descriptors.

→ predict_pKa Takes molecular descriptors as input and returns predicted pKa values using the loaded model.

👩‍💻 Installation

1. Clone the repository:

git clone https://github.com/anastasiafloris/pKaPredict.git
cd pKaPredict

2. Create and activate the conda environnement:

conda create -n pkapredict_env python=3.10 -y
conda activate pkapredict_env

3. Install the package:

pip install pKaPredict

🍏 For macOS users (⚠ required for LightGBM to work):

4. Install the system library libomp:

brew install libomp

If brew is not installed, follow the instructions here: https://brew.sh

5. Navigate to the script folder of the package in your terminal:

cd script

6. The package is yours 🎁:

Run the runpackage.py file in your terminal to predict the pKa of a molecule of your choice, using its SMILES string:

python runpackage.py

🪪 License

This project is licensed under the MIT License.
You are free to use, modify, and distribute it with proper attribution.

📗 References

The dataset used in this project is the test_acids_bases_descfinal_nozwitterions.csv file from the cbio3lab repository.
It was originally extracted from the Harvard Dataverse.

👯‍♀️ Authors

This project was completed as part of the EPFL course Practical Programming in Chemistry.

• Anastasia Floris
• Candice Habert