Lipophilicity Model with Morgan Fingerprints

This repository contains a Python package and script to train a model predicting lipophilicity based on Morgan fingerprints of molecular SMILES representations. The model uses an MLP regressor and evaluates its performance using Root Mean Squared Error (RMSE) on test data.

Contents

• src/train_model.py: Script for loading data, generating fingerprints, training the model, and saving evaluation results.
• data/Lipophilicity.csv: Data file.
• config.json: Model hyperparameters specification.
• environment.yml: Conda environment file listing dependencies.
• results.txt: Output file that stores the model's RMSE, the conda environment used, and key hyperparameters.

Installation

1. Clone the repository:

   git clone https://github.com/kbhalajiyadav/lipophilicity_model.git
   cd lipophilicity_model

2. Set up the environment: Create a new Conda environment using the environment.yml file:

   conda env create -f environment.yml
   conda activate molecule_modeling  # Replace with your actual environment name

Usage

You can specify model hyperparameters either through a JSON configuration file or by using command-line arguments.

Running the Script

To run the model training script with a JSON configuration file:

python src/train_model.py --config config.json

Specifying Hyperparameters Directly in the Command Line

If you prefer, you can specify hyperparameters directly on the command line. For example:

python src/train_model.py --hidden_layer_sizes 100,50 --alpha 0.01

Using Both JSON and Command-Line Arguments

When both are used, command-line arguments will override values from the JSON file:

python src/train_model.py --config config.json --alpha 0.01

Config File Example

Create a JSON configuration file like this in the main directory:

{
   "hidden_layer_sizes": [100, 100],
   "alpha": 0.001
}

Output

The script will save:

• The RMSE for the test set,
• The name of the active conda environment, and
• Hyperparameter settings

...to results.txt in the main directory.

• Arguments:
  • --hidden_layer_sizes: Specifies the architecture of the neural network.
  • --alpha: Sets the regularization strength of the model.

After execution, the script will output the test set RMSE, current environment, and hyperparameters to results.txt.

License

This project is licensed under the Apache-2.0 License.

--