Shehab Eman, Nayel Hamada, Taha Mohamed
Department of Computer Science, Faculty of Computers and Artificial Intelligence, University of Sadat City, Sadat, Egypt.
Department of Computer Science, Faculty of Computers and Artificial Intelligence, Benha University, Benha, Egypt.
J Comput Aided Mol Des. 2025 Apr 26;39(1):21. doi: 10.1007/s10822-025-00597-1.
Predicting molecular toxicity is an important stage in the process of drug discovery. It is directly related to medical destiny and human health. This paper presents an enhanced model for chemical respiratory toxicity prediction. It used a combination of molecular descriptors and term frequency - inverse document frequency (TF-IDF) based models with different machine learning algorithms. To address class imbalance, SMOTE is applied. Appropriate hyper-parameter tuning is required to generate a better system with a classifier. So, we adjusted the hyper-parameters of various models and used the adjusted parameters to train the model. We tuned hyper-parameters using OPTUNA. Internal and external validation were used to confirm the models' performance. According to the results, the model's internal validation accuracy and AUC using the random forest approach were 88.6% and 93.2%. For external validation, the model's accuracy value using random forest and Gradient Boosting Classifier were 92.2% with AUC 97%. Comparing these results with previous studies shows that our model performs better compared to them.
预测分子毒性是药物发现过程中的一个重要阶段。它直接关系到医学命运和人类健康。本文提出了一种用于化学物质呼吸毒性预测的增强模型。该模型结合了分子描述符和基于词频 - 逆文档频率（TF-IDF）的模型，并采用了不同的机器学习算法。为了解决类别不平衡问题，应用了SMOTE算法。需要进行适当的超参数调整，以使用分类器生成更好的系统。因此，我们调整了各种模型的超参数，并使用调整后的参数训练模型。我们使用OPTUNA调整超参数。通过内部和外部验证来确认模型的性能。根据结果，使用随机森林方法时模型的内部验证准确率和AUC分别为88.6%和93.2%。对于外部验证，使用随机森林和梯度提升分类器时模型的准确率值为92.2%，AUC为97%。将这些结果与先前的研究进行比较表明，我们的模型比它们表现更好。
