School of Computer Science and Artificial Intelligence, Changzhou University, Changzhou 213164, China.
School of Computer Science and Artificial Intelligence, Changzhou University, Changzhou 213164, China.
Comput Biol Chem. 2023 Jun;104:107857. doi: 10.1016/j.compbiolchem.2023.107857. Epub 2023 Apr 1.
Microbes in the human body are closely linked to many complex human diseases and are emerging as new drug targets. These microbes play a crucial role in drug development and disease treatment. Traditional methods of biological experiments are not only time-consuming but also costly. Using computational methods to predict microbe-drug associations can effectively complement biological experiments. In this experiment, we constructed heterogeneity networks for drugs, microbes, and diseases using multiple biomedical data sources. Then, we developed a model with matrix factorization and a three-layer heterogeneous network (MFTLHNMDA) to predict potential drug-microbe associations. The probability of microbe-drug association was obtained by a global network-based update algorithm. Finally, the performance of MFTLHNMDA was evaluated in the framework of leave-one-out cross-validation (LOOCV) and 5-fold cross-validation (5-fold CV). The results showed that our model performed better than six state-of-the-art methods that had AUC of 0.9396 and 0.9385 + /- 0.0000, respectively. This case study further confirms the effectiveness of MFTLHNMDA in identifying potential drug-microbe associations and new drug-microbe associations.
人体内的微生物与许多复杂的人类疾病密切相关，它们正成为新的药物靶点。这些微生物在药物开发和疾病治疗中起着至关重要的作用。传统的生物学实验方法不仅耗时，而且成本高昂。使用计算方法预测微生物-药物的关联可以有效地补充生物学实验。在这个实验中，我们使用多种生物医学数据来源构建了药物、微生物和疾病的异质网络。然后，我们开发了一个基于矩阵分解和三层异质网络的模型（MFTLHNMDA）来预测潜在的药物-微生物关联。通过全局网络更新算法获得微生物-药物关联的概率。最后，在留一交叉验证（LOOCV）和 5 折交叉验证（5-fold CV）框架下评估 MFTLHNMDA 的性能。结果表明，我们的模型比六种最先进的方法表现更好，它们的 AUC 分别为 0.9396 和 0.9385+/-0.0000。这个案例研究进一步证实了 MFTLHNMDA 在识别潜在药物-微生物关联和新的药物-微生物关联方面的有效性。
