Evaluation of binary QSAR models derived from LUDI and MOE scoring functions for structure based virtual screening

Area of Science:

• Computational chemistry
• Drug discovery
• cheminformatics

Background:

• Structure-based virtual screening is key for prioritizing new chemical entities (NCEs).
• Scoring function accuracy in predicting binding free energy remains a bottleneck.
• Using scoring functions as filters to distinguish binders from nonbinders is an improved approach.

Purpose of the Study:

• To compare the discriminative ability of binary QSAR models derived from LUDI and MOE scoring functions.
• To evaluate these models against established classification methods.
• To assess the utility of LUDI and MOE scoring functions in virtual screening.

Main Methods:

• Development of binary QSAR models using LUDI and MOE scoring functions.
• Comparison with models from Jacobsson et al. using PLS discriminant analysis, rule-based, and Bayesian classification.
• Testing on five diverse datasets: ERalpha_mimics, ERalpha_toxins, MMP-3, fXa, and AChE.

Main Results:

• Binary QSAR models demonstrated comparable discriminative ability to PLS, rule-based, and Bayesian methods.
• LUDI and MOE scoring functions effectively handle diverse protein-ligand interactions.
• Performance of LUDI and MOE is comparable to other established scoring functions like ICM and Cscore.

Conclusions:

• Binary QSAR models using LUDI and MOE scoring functions are effective for distinguishing binders from nonbinders.
• These models offer a viable alternative for preliminary screening in virtual drug discovery.
• The findings support the integration of these QSAR models into multi-layered virtual screening strategies.