A deep learning framework for predicting infrared (IR) spectra using stereochemical molecular graph representations.
SIMG-IR leverages graph neural networks to predict IR spectra for:
- Single molecules
- Multi-component molecular mixtures
The model uses a specialized stereochemical graph representation that captures 3D molecular structure and connectivity information critical for accurate IR spectrum prediction.
- Stereochemical graph construction from molecular structures
- Graph neural network architecture optimized for spectral prediction
- Support for both individual molecules and molecular mixtures
- High-resolution IR spectra prediction (4 cm⁻¹ resolution)
- Parallel training across multiple GPUs
The pipeline consists of:
- Preprocessing molecular data into stereochemical graphs
- Training the GNN model
- Predicting IR spectra for new molecules
See documentation for detailed usage instructions.