Classification models and SAR analysis on CysLT1 receptor antagonists using machine learning algorithms.

Area of Science:

• Medicinal Chemistry
• Computational Drug Discovery
• Pharmacology

Background:

• The cysteinyl leukotrienes 1 (CysLT1) receptor is a significant therapeutic target for allergic conditions such as rhinitis.
• Developing effective CysLT1 receptor antagonists is crucial for managing these diseases.

Purpose of the Study:

• To construct and evaluate classification models for predicting the bioactivity of CysLT1 receptor antagonists.
• To identify key molecular features and substructures associated with high antagonist activity.

Main Methods:

• A dataset of 503 CysLT1 receptor antagonists was curated and divided into highly active (<1000 nM) and weakly active (≥1000 nM) groups.
• Molecules were characterized using CORINA descriptors, MACCS fingerprints, Morgan fingerprints, and molecular SMILES.
• Classification models were built using random forests (RF), deep neural networks (DNN), and recurrent neural networks (RNN) with self-attention.

Main Results:

• All developed models achieved a test set accuracy of at least 85%.
• The best performing model (Model 2C) demonstrated a high accuracy of 93%.
• Structure-activity relationship (SAR) analyses revealed that highly active antagonists frequently contain tetrazole, indole, and quinoline substructures.

Conclusions:

• Computational models can effectively predict CysLT1 receptor antagonist bioactivity.
• The identified substructures (tetrazoles, indoles, quinolines) are promising pharmacophores for designing more potent antagonists.
• This study provides valuable insights for the rational design of novel drugs targeting CysLT1 receptors for allergic diseases.