Prediction of chemical warfare agents based on cholinergic array type meta-predictors

Area of Science:

• Computational chemistry
• Toxicology
• Drug discovery

Background:

• Molecular insights are crucial for sustainable development and risk assessment, particularly for managing emerging harmful agents like cholinergic chemical warfare agents (CWAs).
• Understanding drug target interactions (DTIs) is key to predicting the effects of chemical agents.

Purpose of the Study:

• To develop robust classification models for predicting drug target interactions (DTIs) related to cholinergic activity.
• To create multi-task meta-predictors capable of both cholinergic prediction and chemical warfare agent (CWA) detection.

Main Methods:

• Molecular structures of known cholinergic agents were encoded using molecular descriptors.
• Drug target interaction (DTI) data and cholinergic activities were used to train classification models for five cholinergic targets.
• Ensemble methods were employed for statistical validation, achieving high performance metrics.
• Classifiers were transformed into 2D or 3D array type meta-predictors for multi-task learning.

Main Results:

• Classification models demonstrated reliable statistical validation with ensemble-AUC up to 0.790, MCC up to 0.991, and accuracy up to 0.995.
• The array-type meta-predictors showed strong detection capabilities for CWAs, even with imbalanced datasets.
• Performance metrics for CWA detection included area under the precision-recall curve up to 0.997, MCC up to 0.638, and F1-scores for non-CWAs and CWAs up to 0.991 and 0.585, respectively.

Conclusions:

• The developed computational models provide a valuable tool for predicting cholinergic activity and detecting chemical warfare agents (CWAs).
• These methods contribute to advancing chemical safety and risk assessment strategies for sustainable development.
• The multi-task learning approach effectively addresses the challenge of identifying harmful agents in complex chemical environments.