mamba

mamba predicts and classifies chemical bond orders (single, double, triple, aromatic) from spatial atomic coordinates derived from quantum chemical calculations or crystallographic data using machine learning for bond perception.

Key Features:

• Machine Learning Approach: Learns rules for novel chemistry and specific functional groups by training on structure-data files (.sd/.sdf) that contain bond information.
• Data Utilization: Uses connection tables and bond blocks from molecular data files to extract chemical knowledge for prediction.
• Spatial Coordinates: Operates on spatial atomic coordinates derived from quantum chemical calculations or crystallographic data.
• Training on Structure Data: Trained on structure-data files (.sd/.sdf) that include bond information for supervised learning of bond types.
• Accuracy: When trained on a sufficiently large and diverse database, can identify chemical bond orders with accuracy comparable to traditional hard-coded cheminformatics rules and algorithms.
• Integration with Python Libraries: Integrates RDKit for cheminformatics, scikit-learn for machine learning algorithms, and pandas for data manipulation.
• Implementation: Implemented in Python.

Scientific Applications:

• Chemical Bond Identification: Identification and classification of bond types in molecular structures for drug discovery, materials science, and computational chemistry.
• Novel Chemistry Exploration: Prediction of bond types in uncharacterized or novel chemical structures to support analysis of new functional groups and chemistries.

Methodology:

Training on structure-data files (.sd/.sdf) containing bond information using machine-learning algorithms with scikit-learn, feature extraction and cheminformatics operations via RDKit, and data handling with pandas, operating on spatial atomic coordinates from quantum chemical calculations or crystallographic data.