FLAP: GRID molecular interaction fields in virtual screening. validation using the DUD data set.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Virtual screening is crucial for identifying potential drug candidates.
• Accurate enrichment of diverse chemical structures (chemotypes) is essential for efficient screening.
• FLAP (Fingerprints for Ligands and Proteins) is a tool used in virtual screening.

Purpose of the Study:

• To assess the performance of FLAP in virtual screening using a diverse dataset.
• To evaluate various ligand- and receptor-based screening approaches and data fusion methods.
• To determine the impact of prior knowledge and decoy data on screening accuracy.

Main Methods:

• Utilized a subset of the Directory of Useful Decoys (DUD) dataset with 13 targets.
• Examined ligand-based and receptor-based screening using templates, 2D structures, and receptor structures.
• Applied data fusion techniques, including Pareto ranking, to combine screening results.
• Incorporated inactivity or decoy data to train FLAP's scoring function.

Main Results:

• Achieved excellent chemotype enrichment (approx. 17-fold over random at 1% false positive rate).
• Demonstrated that increased starting knowledge improves enrichment.
• Showcased up to 50% enrichment improvement using Pareto ranking for data fusion.
• Observed a further 2-fold improvement in recovery when training with decoy data.

Conclusions:

• FLAP shows significant utility for virtual screening, adaptable to varying levels of prior knowledge.
• Data fusion, especially with Pareto ranking, enhances screening efficiency.
• Incorporating decoy data further refines FLAP's predictive power for drug discovery.