CSAR 2014: A Benchmark Exercise Using Unpublished Data from Pharma.

Area of Science:

• Computational chemistry and drug discovery
• Structural bioinformatics and molecular modeling

Background:

• The 2014 Community Structure-Activity Relationship (CSAR) Benchmark Exercise utilized unpublished GlaxoSmithKline (GSK) crystal structures and affinity data.
• Three protein targets were included: tRNA (m1G37) methyltransferase (TrmD), Spleen Tyrosine Kinase (SYK), and Factor Xa (FXa).
• The large dataset size provided statistical significance for method comparisons.

Purpose of the Study:

• To evaluate the performance of various computational methods for molecular docking and scoring.
• To assess the effectiveness of different approaches in predicting protein-ligand poses and ranking binding affinities.
• To identify challenges and successes in computational drug discovery methods using real-world data.

Main Methods:

• Phase 1 involved identifying near-native poses from decoys for given protein-ligand complexes.
• Phase 2 required docking and ranking small molecules based on SMILES strings and protein structures.
• Participants employed diverse methods including empirical, knowledge-based, machine-learning, and shape-fitting approaches.

Main Results:

• Docking performance was generally successful, particularly in Phase 2 compared to Phase 1.
• Scoring and ranking proved more challenging, with no single approach showing a clear advantage.
• Ranking Factor Xa ligands was intractable for all participants; TrmD and SYK ranking saw some success.

Conclusions:

• Decoys complicate analysis and are not recommended for benchmark exercises.
• While docking methods showed promise, robust scoring and ranking remain significant challenges in computational drug discovery.
• Further investigation across more systems is needed to definitively identify superior docking and scoring approaches.