Benchmarking 2D/3D/MD-QSAR Models for Imatinib Derivatives: How Far Can We Predict?

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Imatinib is a key drug for Chronic Myeloid Leukemia (CML), but drug resistance arises from ABL kinase domain mutations.
• Discovering novel bioactive analogues is crucial to overcome Imatinib resistance.
• Quantitative Structure-Activity Relationship (QSAR) models and molecular docking accelerate drug discovery by screening chemical libraries.

Purpose of the Study:

• To develop reliable QSAR models for predicting the binding affinity and inhibition potencies of Imatinib derivatives.
• To explore the utility of 3D and molecular dynamics (MD) descriptors in augmenting traditional 2D QSAR models.
• To identify key dynamic protein-ligand interactions for optimizing future drug design.

Main Methods:

• Employed molecular docking and molecular dynamics (MD) simulations on a large series of Imatinib derivatives.
• Developed an ensemble of QSAR models using deep neural nets (DNN) and hybrid sets of 2D, 3D, and MD descriptors.
• Validated models using rigorous external test sets and 10-fold cross-validation (native and nested).

Main Results:

• DNN regression models demonstrated excellent external prediction performance for pKi (R² ≥ 0.71) and pIC50 (R² ≥ 0.54) datasets.
• Both DNN and random forest models showed similar performance across different descriptor sets.
• Incorporating 3D/MD descriptors did not significantly improve R² but reduced the Mean Absolute Error (MAE) in DNN models.

Conclusions:

• The developed QSAR models provide a reliable framework for predicting the activity of Imatinib derivatives.
• While 3D/MD descriptors did not substantially boost R², they improved MAE and offered insights into dynamic interactions.
• These models can guide the design of novel CML therapeutics with improved efficacy and resistance profiles.