Department of Electronics and Information Engineering, Jeonbuk National University, Jeonju 54896, Korea.
Institute of Avionics and Aeronautics (IAA), Air University, Islamabad 44000, Pakistan.
Int J Mol Sci. 2022 Jul 27;23(15):8314. doi: 10.3390/ijms23158314.
N6-methyladenine (6mA) has been recognized as a key epigenetic alteration that affects a variety of biological activities. Precise prediction of 6mA modification sites is essential for understanding the logical consistency of biological activity. There are various experimental methods for identifying 6mA modification sites, but in silico prediction has emerged as a potential option due to the very high cost and labor-intensive nature of experimental procedures. Taking this into consideration, developing an efficient and accurate model for identifying N6-methyladenine is one of the top objectives in the field of bioinformatics. Therefore, we have created an in silico model for the classification of 6mA modifications in plant genomes. ENet-6mA uses three encoding methods, including one-hot, nucleotide chemical properties (NCP), and electron-ion interaction potential (EIIP), which are concatenated and fed as input to ElasticNet for feature reduction, and then the optimized features are given directly to the neural network to get classified. We used a benchmark dataset of rice for five-fold cross-validation testing and three other datasets from plant genomes for cross-species testing purposes. The results show that the model can predict the N6-methyladenine sites very well, even cross-species. Additionally, we separated the datasets into different ratios and calculated the performance using the area under the precision-recall curve (AUPRC), achieving 0.81, 0.79, and 0.50 with 1:10 (positive:negative) samples for , , and , respectively.
N6-甲基腺嘌呤（6mA）已被认为是影响多种生物活性的关键表观遗传修饰。准确预测 6mA 修饰位点对于理解生物活性的逻辑一致性至关重要。有各种用于识别 6mA 修饰位点的实验方法，但由于实验程序的成本非常高且劳动强度大，因此基于计算机的预测已成为一种潜在选择。考虑到这一点，开发一种用于识别 N6-甲基腺嘌呤的高效且准确的模型是生物信息学领域的首要目标之一。因此，我们创建了一个用于植物基因组中 6mA 修饰分类的计算机模型。ENet-6mA 使用三种编码方法，包括 one-hot、核苷酸化学性质（NCP）和电子-离子相互作用势能（EIIP），将它们串联并作为输入提供给弹性网络进行特征减少，然后将优化的特征直接提供给神经网络进行分类。我们使用水稻的基准数据集进行五折交叉验证测试，并使用来自植物基因组的另外三个数据集进行跨物种测试。结果表明，该模型可以很好地预测 N6-甲基腺嘌呤位点，甚至可以进行跨物种预测。此外，我们将数据集分为不同的比例，并使用精度-召回曲线下的面积（AUPRC）计算性能，对于 、 、 和 ，当正样本：负样本分别为 1:10 时，AUPRC 分别达到 0.81、0.79、0.50。
