Integrating QSAR modelling with reinforcement learning for Syk inhibitor discovery

Area of Science:

• Medicinal Chemistry
• Computational Drug Discovery
• Pharmacology

Background:

• Spleen tyrosine kinase (Syk) is a key mediator in inflammatory pathways and a therapeutic target for autoimmune disorders like immune thrombocytopenia.
• Existing Syk inhibitors exhibit suboptimal efficacy and safety, driving the need for novel drug discovery approaches.

Purpose of the Study:

• To develop and apply a novel deep reinforcement learning strategy for the identification of new spleen tyrosine kinase (Syk) inhibitors.
• To generate structurally novel Syk inhibitors with high predicted potency and favorable drug-like properties.

Main Methods:

• Integration of quantitative structure-activity relationship (QSAR) predictions with generative modeling using a stacking-ensemble approach.
• Generation and screening of over 78,000 molecules to identify promising Syk inhibitor candidates.
• Evaluation of candidates based on predicted potency, binding affinity, and drug-likeness.

Main Results:

• A deep reinforcement learning model achieved a correlation coefficient of 0.78 in predicting molecular activity.
• Identified 139 candidate molecules with high predicted potency, binding affinity, and drug-likeness.
• The identified compounds demonstrated structural novelty while retaining key Syk inhibitor characteristics.

Conclusions:

• The developed QSAR-guided reinforcement learning framework accelerates the discovery of novel Syk inhibitors.
• This methodology offers a versatile approach for drug discovery, particularly for rare disease therapeutics.
• The study presents the first application of QSAR-guided reinforcement learning for Syk inhibitor discovery, yielding promising novel candidates.