Multinomial classification of NLRP3 inhibitory compounds based on large scale machine learning approaches.

Area of Science:

• Immunology
• Computational Chemistry
• Pharmacology

Background:

• The NLRP3 inflammasome plays a critical role in innate immunity and is implicated in various diseases, including autoimmune and inflammatory conditions.
• Machine learning (ML) is increasingly utilized in pharmaceutical research for drug discovery and development.
• Identifying NLRP3 inhibitors is essential for therapeutic interventions targeting NLRP3-mediated diseases.

Purpose of the Study:

• To apply machine learning approaches for the multinomial classification of NLRP3 inhibitors.
• To evaluate the impact of data imbalance and the effectiveness of the synthetic minority oversampling technique (SMOTE) in improving classification accuracy.
• To develop and validate quantitative structure-activity relationship (QSAR) models for predicting NLRP3 inhibitors.

Main Methods:

• Quantitative structure-activity relationship (QSAR) modeling was performed using 154 molecules from the ChEMBL database (version 29).
• Multinomial classification models were developed, with a focus on addressing data imbalance using SMOTE.
• Model performance was assessed using metrics such as accuracy and log loss, and receiver operating characteristic (ROC) curves were analyzed.

Main Results:

• The developed QSAR classification models achieved high accuracy, with top models ranging from 0.86 to 0.99.
• SMOTE significantly improved the sensitivity of classifiers to minority groups and enhanced overall model accuracy.
• Tuning parameters and handling imbalanced data led to substantial improvements in ROC plot values.

Conclusions:

• The QSAR classification models demonstrated robust statistical performance and interpretability.
• SMOTE is an effective technique for handling imbalanced datasets in ML-based drug discovery.
• These models show strong potential for the rapid screening of novel NLRP3 inhibitors.