This repository explores and compares different metrics for evaluating ligands. Ligand validation is essential in structural biology and drug discovery — and while the Q-score is widely used, it has limitations, especially with small molecules.
To improve evaluation robustness, we test and analyze several metrics:
- Q-score – the current standard, but sensitive to resolution and background noise.
- Wasserstein Distance (WD) – spatially-aware and effective in detecting structural changes.
- Total Variation Distance (TVD) – efficient but sometimes too insensitive to small changes.
🧪 We evaluate how each metric performs in various conditions:
- Symmetry and consistency
- Sensitivity to added/missing density
- Positional shifts of ligands
- Background noise robustness
Both synthetic and real ligand data are used in the experiments to assess how well each metric captures structural similarity.
Authors: Martyna Stasiak, Maria Musiał, Patryk Janiak, Mateusz Bernart
Supervisor: Dariusz Brzeziński
Affiliation: Poznań University of Technology (2024/25, Winter Semester)