Gu Wei-Cheng, Ma Bin-Guang
Hubei Key Laboratory of Agricultural Bioinformatics, College of Informatics, Huazhong Agricultural University, Wuhan, 430070, China.
BMC Genomics. 2025 Aug 1;26(1):712. doi: 10.1186/s12864-025-11863-9.
Predicting bacterial transcriptional regulatory networks (TRNs) through computational methods is a core challenge in systems biology, and there is still a long way to go. Here we propose a powerful, general, and stable computational framework called PGBTR (Powerful and General Bacterial Transcriptional Regulatory networks inference method), which employs Convolutional Neural Networks (CNN) to predict bacterial transcriptional regulatory relationships from gene expression data and genomic information. PGBTR consists of two main components: the input generation step PDGD (Probability Distribution and Graph Distance) and the deep learning model CNNBTR (Convolutional Neural Networks for Bacterial Transcriptional Regulation inference). On the real Escherichia coli and Bacillus subtilis datasets, PGBTR outperforms other advanced supervised and unsupervised learning methods in terms of AUROC (Area Under the Receiver Operating Characteristic Curve), AUPR (Area Under Precision-Recall Curve), and F1-score. Moreover, PGBTR exhibits greater stability in identifying real transcriptional regulatory interactions compared to existing methods. PGBTR provides a new software tool for bacterial TRNs inference, and its core ideas can be further extended to other molecular network inference tasks and other biological problems using gene expression data.
通过计算方法预测细菌转录调控网络(TRNs)是系统生物学中的一项核心挑战,仍有很长的路要走。在此,我们提出了一个强大、通用且稳定的计算框架,称为PGBTR(强大且通用的细菌转录调控网络推理方法),它采用卷积神经网络(CNN)从基因表达数据和基因组信息中预测细菌转录调控关系。PGBTR由两个主要部分组成:输入生成步骤PDGD(概率分布和图距离)和深度学习模型CNNBTR(用于细菌转录调控推理的卷积神经网络)。在真实的大肠杆菌和枯草芽孢杆菌数据集上,PGBTR在受试者工作特征曲线下面积(AUROC)、精确率-召回率曲线下面积(AUPR)和F1分数方面优于其他先进的监督和无监督学习方法。此外,与现有方法相比,PGBTR在识别真实转录调控相互作用方面表现出更高的稳定性。PGBTR为细菌TRNs推理提供了一种新的软件工具,其核心思想可以进一步扩展到使用基因表达数据的其他分子网络推理任务和其他生物学问题。