Hybrid QSAR modeling of LD50 in organophosphorus nerve agents: a mechanistic approach using DFT and molecular docking

Chemical warfare agents (CWAs), particularly organophosphorus (OP) nerve agents, are among the most toxic and persistent compounds known, posing significant threats to human health and security. Experimental determination of their median lethal dose (LD₅₀) values is limited by ethical, biosafety, and accessibility constraints. While conventional QSAR models provide useful approximations, they often lack mechanistic interpretability, especially for novel agents.

In this study, we present a hybrid QSAR framework that integrates mechanistically relevant descriptors derived from density functional theory (DFT) and molecular docking simulations with conventional physicochemical features to predict LD₅₀ of OP nerve agents. The key mechanistic descriptors include acetylcholinesterase (AChE) binding affinity and serine phosphorylation interaction energy, capturing distinct toxicodynamic phases of nerve agent action.

We evaluate both linear regression and random forest models to assess predictive performance and interpretability. Cross-validation confirms that incorporating mechanistic features modestly improves accuracy and generalizability. Feature importance analysis identifies interaction energy as the most influential predictor, aligning with the irreversible inhibition mechanism of AChE.

Importantly, the model is capable of predicting LD₅₀ values for structurally untested agents, including GF and Novichok compounds, thereby extending its utility to substances lacking experimental data. This study highlights the potential of mechanistically grounded in silico methods as an ethically sound and scalable alternative to animal testing for acute toxicity assessment. By aligning with regulatory needs for interpretable and reproducible predictions, the proposed approach contributes to integrated testing strategies, and new approach methodologies in computational toxicology.