Related Experiment Videos
Lead hopping using SVM and 3D pharmacophore fingerprints
Jamal C Saeh1, Paul D Lyne, Bryan K Takasaki
1Cancer Discovery, AstraZeneca R and D Boston, 35 Gatehouse Drive, Waltham, Massachusetts 02451, USA. jamal.saeh@astrazeneca.com
Journal of Chemical Information and Modeling
|July 28, 2005
Summary
This study developed predictive models using 3D pharmacophore fingerprints and support vector machines to classify G-protein-coupled receptor (GPCR) assay compounds. The models accurately identified active and inactive compounds, even with novel chemical structures.
Area of Science:
- Computational chemistry
- Cheminformatics
- Drug discovery
Background:
- G-protein-coupled receptors (GPCRs) are crucial drug targets.
- Accurate classification of compound activity is vital for drug discovery.
- Predictive modeling aids in identifying potential drug candidates.
Purpose of the Study:
- To develop robust classification models for GPCR assays.
- To assess model performance on chemically distinct validation sets.
- To simulate lead-hopping scenarios for predictive model evaluation.
Main Methods:
- Utilized 3D pharmacophore fingerprints for molecular representation.
- Employed support vector machine (SVM) classification algorithm.
- Designed challenging validation sets distinct from training data.
- Simulated lead-hopping by excluding specific compound classes from training.
Main Results:
- Generated accurate models for classifying active/inactive compounds in GPCR assays.
- Achieved high accuracy (>99%) on a large set of inactive compounds.
- Successfully recalled 75% of active compounds in a simulated lead-hopping experiment.
- Demonstrated model robustness against novel chemical scaffolds.
Conclusions:
- 3D pharmacophore fingerprints and SVMs provide effective predictive models for GPCRs.
- The models exhibit strong generalization capabilities, even for novel chemical series.
- This approach is valuable for virtual screening and lead optimization in drug discovery.