This research project explores the application of physics-informed machine learning techniques to accurately predict hydration free energies for small molecules. The methodology combines data-driven approaches with physics-based insights to enhance predictive capabilities.
eda_pp.ipynb: Conducts exploratory data analysis (EDA) and preprocessing, using visualization tools likematplotlibandseabornto understand molecular datasets.make_groups.ipynb: Organizes molecules into meaningful categories, leveraging structural or chemical properties to aid in model training.regression.ipynb: Implements and evaluates regression models tailored to predict hydration free energies, using metrics like mean absolute error (MAE), mean squared error (MSE), and R-squared.
- Physics-informed EDA: Utilizes domain knowledge for effective data preparation and analysis.
- Data Structuring: Tools for categorizing molecules based on features critical to hydration energy predictions.
- Advanced Regression Models: Machine learning models that integrate physics-based principles to improve accuracy and reliability.
Ensure the following are installed. It is highly recommended to use the provided environment.yml file to create a consistent environment.
- Python 3.12.2
- The following libraries, with the specified versions:
numpy==1.26.4scipy==1.15.0pandas==2.2.3scikit-learn==1.6.1xgboost==2.1.3lightgbm==4.5.0plotly==5.24.1matplotlib==3.10.0fastapi==0.115.7
-
Using
conda:Conda installation using miniforge -
curl -L -O "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh" bash Miniforge3-$(uname)-$(uname -m).sh
Windows - Download Miniforge3 for Windows (64-bit)
It is highly recommended to use
condato create an environment with the specified dependencies. This ensures that you have the correct versions of all the required packages.-
Create the environment from the
environment.ymlfile:conda env create -f environment.yml
-
Activate the environment:
conda activate scis
-
-
Alternative: Using
pip(Not Recommended):While not recommended, you can attempt to install the dependencies using
pip. However, this may lead to version conflicts or other issues.-
Create a
requirements.txtfile with the following content:numpy==1.26.4 scipy==1.15.0 pandas==2.2.3 scikit-learn==1.6.1 xgboost==2.1.3 lightgbm==4.5.0 plotly==5.24.1 matplotlib==3.10.0 fastapi==0.115.7
-
Install the dependencies:
pip install -r requirements.txt
-
-
EDA and Preprocessing:
Open
eda_pp.ipynbto:- Visualize molecular data distributions.
- Perform feature engineering and data preprocessing.
-
Data Grouping:
Use
make_groups.ipynbfor:- Organizing molecules based on their physical or chemical properties.
- Structuring data to enhance model training and evaluation.
-
Regression Modeling:
Run
regression.ipynbto:- Train machine learning models on hydration free energy datasets.
- Evaluate model performance using key metrics.
Contributions to this research are welcome. Suggestions, code improvements, and additional experiments can be submitted via pull requests.
Marvin V Prakash, Ajeet Yadav, Dr. Pradipta Bandyopadhyay
Add the relevant links here, such as:
- https://github.com/MobleyLab/FreeSolv
- Link to the GitHub repository.
- Links to any relevant datasets or external resources.
- Links to documentation for the specific libraries used.