Boosting Predictability: Towards Rapid Estimation of Organic Molecule Solubility

Machine Learning for Predicting Solubility

The water solubility of organic molecules is critical for optimizing the performance and stability of aqueous flow batteries, as well as for various other applications. While solubility measurements are relatively straightforward in some cases, theoretical predictions remain a significant challenge. Machine learning algorithms have become invaluable tools over the past decade to address this. High-quality data and effective descriptors are essential for building reliable, data-driven estimation models. This repository systematically investigates the effectiveness of enhanced structure-based descriptors and an outlier detection procedure to improve aqueous solubility predictability.

1. Clone the repository:

   git clone git@github.com:sahashemip/ML4OrganicMoleculeSolubility.git

    
2. Navigate to the project directory:

   cd ML4OrganicMoleculeSolubility

    
3. Install the required dependencies:

   pip install -r requirements.txt
   
   

   How to Use

   1. Navigate to the notebooks directory:

      • Open and run the Jupyter notebooks sequentially based on the numbering:
        1. analysis
        2. descriptors
        3. ml-models
        4. outlier-detection

   2. Outlier Detection:

      • To perform outlier detection, modify the parameters in the outlier_detector.py script. Refer to the data in TABLE I of the associated manuscript for parameter details.

   Project Structure

   • notebooks/: Contains step-by-step Jupyter notebooks for analysis, feature engineering, and model development.
   • scripts/: Includes Python scripts for outlier detection and custom preprocessing utilities.
   • datasets/: Holdes all different datasets generated by distinct descriptors.
   • outliers/: Stores outputs related to the detected outliers.