Extrapolative prediction using physically-based QSAR.

Area of Science:

• Computational chemistry
• Drug discovery
• Structure-activity relationship studies

Background:

• Quantitative Structure-Activity Relationship (QSAR) models are crucial for predicting drug efficacy.
• Accurate ligand pose determination is a significant challenge in QSAR modeling.
• Existing methods often rely on manual molecular alignments, limiting their applicability.

Purpose of the Study:

• To evaluate the performance and applicability of the Surflex-QMOD model for binding affinity prediction.
• To demonstrate QMOD's capability in handling diverse chemical structures and targets.
• To compare QMOD's automated alignment approach with traditional manual alignment methods.

Main Methods:

• Application of the QMOD model to a 3D-QSAR benchmark dataset.
• Automated flexible molecular alignment for testing new ligands.
• Comparison of QMOD performance against CoMFA, CoMSIA, and CMF methods.
• Validation using a large, structurally diverse dataset from ChEMBL.

Main Results:

• QMOD demonstrated comparable performance to established methods on a limited test set.
• The model successfully extrapolated predictions to a large, diverse dataset of novel ligands.
• Statistically significant predictions were achieved for ligands synthesized years after model construction.
• QMOD identified potent and structurally novel ligands with high predicted activity (pK(i) ≥ 7.5).

Conclusions:

• Surflex-QMOD effectively integrates structural and activity data for robust binding affinity prediction.
• The automated ligand pose determination addresses a key limitation in traditional QSAR.
• QMOD shows broad applicability and potential for identifying novel drug candidates across diverse targets.