Keras/TensorFlow in Drug Design for Immunity Disorders

Area of Science:

• Computational chemistry and cheminformatics
• Immunology and pharmacology
• Artificial intelligence in drug discovery

Background:

• Immune system homeostasis relies on white blood cells and cytokine receptors.
• Chemokines and their receptors mediate immune cell movement in health and disease.
• Inflammatory disorders necessitate novel, effective therapeutic agents.

Purpose of the Study:

• To discover novel compound scaffolds targeting chemokine receptors CCR2, CCR3, and CXCR3.
• To utilize a Keras/TensorFlow neural network (NN) as a multi-class classifier for compound screening.
• To enhance drug discovery by improving binding affinity and predicting receptor subtype selectivity.

Main Methods:

• Structure-based virtual screening (SBVS) combined with Keras/TensorFlow NN.
• All-atom molecular dynamics simulations to assess binding affinity.
• Comparative analysis of predicted compounds against known receptor antagonists.

Main Results:

• Identification of novel compound scaffolds with potential activity against CCR2, CCR3, and CXCR3.
• Demonstration of NN's efficacy in classifying compound activity and predicting receptor subtype selectivity.
• Achieved high accuracy in receptor subtype prediction, outperforming traditional SBVS for selectivity.

Conclusions:

• Keras/TensorFlow NN offers significant advantages in drug discovery, complementing SBVS.
• The developed NN models accurately predict receptor subtype selectivity, a challenge for SBVS.
• NN models show potential for identifying targeted therapies for immune-related disorders.