Dataset and Code: Formulation-Aware QSAR for Inhaled Neuraminidase Inhibitors with Explainable AI

This repository contains the curated dataset, trained model predictions, analysis code, and publication-quality figures for the manuscript: Noorizadehtehrani, S.; Formulation-Aware QSAR for Inhaled Neuraminidase Inhibitors: Integrating Aerodynamic Descriptors into Predictive Models with Explainable Artificial Intelligence. J. Chem. Inf. Model. 2026. Contents: Curated ChEMBL dataset (1,944 neuraminidase inhibitors with 30 molecular descriptors) Model predictions (Random Forest, Gradient Boosting, SVR) QSAR training code and figure generation scripts All manuscript figures (PNG + TIFF) Best model: Random Forest (R² = 0.550, RMSE = 1.007, 5-fold CV R² = 0.589 ± 0.036)本仓库包含对应稿件的精心整理数据集、训练好的模型预测结果、分析代码以及出版级高质量图表： Noorizadehtehrani, S.；《面向吸入型神经氨酸酶抑制剂的配方感知型定量构效关系：结合空气动力学描述符与可解释人工智能构建预测模型》，发表于《化学信息与建模杂志》（*Journal of Chemical Information and Modeling*）2026年。 开源内容如下： 1. 经整理的ChEMBL数据集（含1944个神经氨酸酶抑制剂，共30种分子描述符） 2. 模型预测结果（涵盖随机森林（Random Forest）、梯度提升（Gradient Boosting）、支持向量回归（Support Vector Regression，SVR）） 3. 定量构效关系（Quantitative Structure-Activity Relationship，QSAR）训练代码与图表生成脚本 4. 全部稿件配图（PNG及TIFF格式） 最优模型为随机森林（决定系数$R^2=0.550$，均方根误差RMSE=1.007，5折交叉验证$R^2=0.589 pm 0.036$）