Genetic Algorithm for Frustrated Lewis Pairs (GA-FLP)

Project Description

This project implements a Genetic Algorithm (GA) to optimize Frustrated Lewis Pairs (FLPs) for chemical applications. It leverages computational chemistry tools like RDKit and xtb for molecular property calculations and a pre-trained model for frustration prediction. The GA explores a chemical space defined by SMILES strings to discover novel FLP structures with desired characteristics.

Setup

Prerequisites

• Python 3.9 or higher
• conda or pip for package management

Installation

git clone https://github.com/lcmd-epfl/ga_flp
conda create -n ga_flp python=3.9 # or your preferred Python version
conda activate ga_flp

# Clone the SCScore repository
git clone https://github.com/connorcoley/scscore.git scscore

pip install -r requirements.txt

git clone https://github.com/lcmd-epfl/NaviCatGA/
conda install xtb==22.1
cd NaviCatGA
python setup.py install
cd ..
rm -rf NaviCatGA

Usage

To run the genetic algorithm, execute the launcher.py script. GA parameters, random seed, and the data file path are now configured via a JSON file.

python launcher.py --config config/config.json

Configuration File (config/config.json)

The config/config.json file contains all the configurable parameters for the Genetic Algorithm run. An example structure is shown below:

{
    "GA_PARAMETERS": {
        "N_GENES": 8,
        "POP_SIZE": 50,
        "MUTATION_RATE": 0.25,
        "N_CROSSOVER_POINTS": 1,
        "NUM_CYCLES": 10,
        "RANDOM_SEED": 42,
        "DATA_FILE": "data/database_HC_bbr.csv"
    }
}

Parameters:

• N_GENES: Number of genes in each chromosome.
• POP_SIZE: Size of the population for the genetic algorithm.
• MUTATION_RATE: Probability of mutation for each gene.
• N_CROSSOVER_POINTS: Number of crossover points during genetic recombination.
• NUM_CYCLES: Number of genetic algorithm cycles to run.
• RANDOM_SEED: Seed for the random number generator to ensure reproducibility.
• DATA_FILE: Path to the database file (CSV or Excel) containing molecular data.

The script will perform a series of GA cycles, optimizing the FLP structures.

Code Structure

• launcher.py: The main script that orchestrates the genetic algorithm. It sets up the GA solver, runs the optimization, and handles logging and result saving.
• ga_core/: Contains core GA-FLP functionalities.
  • ga_core/ga_flp.py: Defines the chromosome_to_smiles and overall_fitness_function used by the GA.
  • ga_core/angle_measure.py: Provides functions for calculating angles in molecular structures.
  • ga_core/ga_flp_fs.py: Contains functions for molecular embedding, property calculations (B-N distance, HA, PA), and xtb optimizations.
  • ga_core/frustration_c.py: Implements the FrustrationPredictor class for predicting molecular frustration using a pre-trained model. See the "Data and Model" section for more details.

Data and Model

This project includes a pre-trained model and the dataset used to train it.

• models/classifier_trained.sav: This is the pre-trained machine learning model (a scikit-learn classifier) used by the FrustrationPredictor for quick predictions of molecular frustration.
• data/selected_data_energies_distances.csv: This CSV file contains the dataset used to train the frustration prediction model. It includes various molecular properties and the "Frustrated" class label, which is used to train the classifier.
• data/quenching/: This directory contains the dataset for training the classifier quenching model.
  • data/quenching/bvs_energies_distances.csv: A CSV file with energies, distances, and other molecular descriptors for a set of molecules.
  • data/quenching/xyz/: A directory containing the corresponding molecular geometry .xyz files for the entries in the CSV.
• xyzs-chtf-profs.zip: Bundle of .xyz files containing optimized geometries of intermediates (I1, I2, I3, I4) and transition states (TS1, TS2) along the catalytic CO$_2$ hydrogenation free-energy profiles of the IFLP catalysts discussed in the manuscript.

Logging

The launcher.py script uses Python's logging module with colored output for better readability. Key information for each GA cycle, such as max fitness, mean fitness, and the best individual's SMILES and chromosome, will be printed to the console.

Troubleshooting

• InconsistentVersionWarning from sklearn: This warning indicates that the classifier_trained.sav model was trained with a different version of scikit-learn than the one currently installed. While it might still work, it's recommended to retrain the model with your current scikit-learn version if you encounter issues or desire optimal performance.
• Warning: Could not load frustration model from classifier_trained.sav. Frustration prediction will be skipped.: This means the classifier_trained.sav file was not found or could not be loaded. Ensure the file is in the correct directory or that its path is correctly specified. If frustration prediction is critical, you might need to train and save the model first.
• TypeError: 'float' object is not subscriptable: This issue has been addressed in ga_core/ga_flp.py. Ensure you have the latest version of the code.
• Excessive "Randomizing starting chromosome." messages: These messages have been suppressed by adjusting the logging level in the navicatGA library.
• Program times out: The GA can be computationally intensive. If the program consistently times out (e.g., with timeout 60), consider increasing the timeout duration or reducing the NUM_CYCLES in launcher.py for quicker runs.

Citing

If you use this code in your research, please cite the following publication:

Das S, Laplaza R, Worakul T, Corminboeuf C. Inverse Design of Frustrated Lewis Pairs for Direct Catalytic CO2 Hydrogenation: Refining and Expanding Design Rules. ChemRxiv. 2025; doi:10.26434/chemrxiv-2025-439dn