solv_strc

Molecular Geometry Prediction Neural Network

A neural network model for predicting molecular geometries in solvent phase given gas phase geometries and solvent dielectric constants.

Overview

1. Load geometry from HDF5 files, create and split dataset
2. Calculate molecular features (global descriptors)
3. Model (MLP)
4. Training function
5. Hyperparameter optimization
6. Prediction
7. Analysis and visualization

Requirements

torch >= 1.9.0
numpy >= 1.21.0
pandas >= 1.3.0
h5py >= 3.1.0
scikit-learn >= 0.24.0
matplotlib >= 3.4.0

Usage

Step 1: Generate Dataset

Generate a dataset from your HDF5 files:

# Generate full dataset
python generate_dataset.py --gas AQM-gas.hdf5 --sol AQM-sol.hdf5 --output molecule_dataset.csv

# Or use `--n-samples N` to generate a sample dataset for testing (N molecules)

Step 2: Train the Neural Network

# training
python train.py --data-path molecule_dataset.csv --epochs 1000

# training with hyperparam optimization
python train.py --data-path molecule_dataset.csv --epochs 1000 --optimize --n-trials 10

Using Pre-trained Model

# Load best model
checkpoint = torch.load('best_model.pt')
model.load_state_dict(checkpoint['model_state_dict'])

# Make prediction (example)
gas_xyz = np.array([[0, 0, 0], [1.5, 0, 0], [0, 1.5, 0]])
atom_nums = np.array([6, 1, 1])
predicted_sol_xyz = predict_geometry(
    model, gas_xyz, atom_nums, dielectric=78.4,
    scaler=data['scaler']
)

print("Predicted solvent geometry:")
print(predicted_sol_xyz)

Output Metrics

• Training/Validation Loss: MSE between predicted and true coordinates
• Test Metrics: MSE, RMSE, MAE, per-coordinate error analysis, per-molecule RMSE distribution