LUNA

LUNA computes hashed protein–ligand interaction fingerprints to improve molecular representations for interpretable machine learning in structure-based virtual screening and drug discovery.

Key Features:

• Implementation: Implemented as a Python 3 toolkit.
• Interaction fingerprints: Introduces EIFP (Extended Interaction FingerPrint), FIFP (Functional Interaction FingerPrint), and HIFP (Hybrid Interaction FingerPrint) to capture comprehensive, functional, and hybrid protein–ligand interaction information, respectively.
• Hashed ECFP-inspired encoding: Encodes protein–ligand interactions into hashed fingerprints inspired by the Extended Connectivity FingerPrint (ECFP) framework.
• Interpretability: Provides interpretability through interaction fingerprints and visualization strategies that map fingerprint features to specific protein–ligand interactions.
• Machine learning integration: Supports development of interpretable machine learning models for structure-based virtual screening.

Scientific Applications:

• Structure-based virtual screening: Improves molecular representation for virtual screening campaigns in drug discovery.
• Docking score reproduction: Demonstrated ability to reproduce DOCK3.7 scores on a dataset of one million docked Dopamine D4 complexes.
• Complex similarity detection: Identifies similarities across molecular complexes that traditional fingerprints may miss.

Methodology:

Calculates and encodes protein–ligand interactions into hashed fingerprints inspired by the Extended Connectivity FingerPrint (ECFP) framework, producing EIFP, FIFP, and HIFP representations.