wsppchem 0.4

Water Solubility Prediction Project

Water Solubility Prediction Project

Overview

This project aims to predict the water solubility of chemical compounds using machine learning techniques. The project developed here can be used to estimate the solubility of new compounds only using the SMILES code of the compounds, which can be valuable in various industries such as pharmaceuticals, agriculture, and environmental science. In this repository, we are making available the data we used to train and test our models and .pkl files containing the optimized parameters of our best model. But more importantly a notebook tracing what we did from the beginning to the end of this project and a package that can predict the water solubility of several SMILEs and of a .csv file containing several SMILEs.

Project Structure

The project is structured as follows:

First, a Notebook containing:

• Import Relevant Modules and Libraries
• Data Collection
• Data Cleaning
• Calculation of RDkit Molecular Descriptors
• Select Machine Learning Models
• Fine-tuning
• Analysis of different models
• Saving of the best trained model and standard scaler

Second, a Package of two main functions containing:

• A function tp predict the LogS value for one or more SMILES
• A function to predicts LogS values for SMILES codes stored in a CSV file

Installation

1. Clone this repository:

git clone https://github.com/Nohalyan/Projetppchem

2. Open your terminal or Anaconda Prompt and navigate to the directory /src/Projectppchem containing the ppchem_environment.yml file and run the following command to create the Conda environment:

conda env create -f ppchem_environment.yml

3. Activate the newly created Conda environment:

conda activate ppchem_environment 

Usage

For the Notebook:

1. Data Preparation: Place your dataset in the data/ directory. Ensure the dataset is formatted correctly with features and labels.
2. Exploratory Data Analysis: Explore the dataset using the Colab notebooks in the notebooks/ directory to understand the data distribution and relationships.
3. Model Training: Use the scripts in the Colab notebooks in the notebooks/ directory to preprocess the data, train machine learning models, and save the trained models in the corresponding models/ directory.
4. Model Evaluation: Evaluate the model performance using the evaluation using the scripts in the Colab notebooks in the notebooks/ directory.
5. Prediction: Once trained, the models in the models/ directory can be used to predict the water solubility of new compounds by providing the required input features.

For the Package:

First, clone our repos If you are using a notebook without the environment, you can download the necessary libraries:

!pip install pandas numpy rdkit tqdm lightgbm

Once the repository has been cloned, you can use the following function to import the functions of our pacakge:

from Projectppchem.src.WSPP import wspp_functions as wspp

The two main functions of our package are predict_logS_smiles and predict_logS_csv which can be used in the following way:

wspp.predict_logS_smiles(*smiles_codes) and wspp.predict_logS_csv(csv_file_path)

The first function wspp.predict_logS_smiles(*smiles_codes) can be used to predict the LogS value for one or more SMILES at the same time. The second fucntion wspp.predict_logS_csv(csv_file_path) can be used to predicts LogS values for SMILES codes stored in a CSV file. And if you need any help, you can use the function wspp.help() which will give you more precise information on the functions as well as an example of how to use them.

License

This project is licensed under the MIT License.

References

This project is based on the code of this Github Jupyter notebook: https://github.com/gashawmg, as well as data from https://github.com/PatWalters.

Authors

• Cossard Lucas: https://github.com/Nohalyan
• Venancio Enzo: https://github.com/Enzo-vnc

This project was carried out as part of EPFL's Practical programming in Chemistry course.