【摘要】 基因调控是生命体中重要的运行机制,运用深度学习来预测p53相关基因的调控关系对了解基因调控具有重要意义。提出了一种基于卷积神经网络与全连接网络相结合的模型。在ArrayExpress获得了电离辐射Affymetrix数据集(E-MEXP-549)上,可通过微阵列表达水平预测基因间调控关系,为保留的验证集提供92.07%分类准确率,并且该模型的kappa系数达到0.84,AUC验证平均精度达到94.25%,从而构建了带有时延性的p53相关的基因调控网络。研究结果表明,该模型在筛选出已经论证调控关系的基因对上具有较好的验证关系,构建出可视化的基因调控网络。在筛选出未知调控关系的基因对上,具有较好的预测关系与研究价值。更多还原

【Abstract】 Gene regulation is an important operating mechanism in living organisms.Deep learning to predict the regulatory relationship of p53 related genes is of great significance for understanding gene regulation.A model based on the combination of CNN and DNN is proposed.ArrayExpress obtained the ionizing radiation Affymetrix data set(E-MEXP-549),which can predict the regulatory relationship between genes through microarray expression levels,providing 92.07%classification accuracy for the retained validation set,and the kappa coefficient of the model reaches 0.84,the average accuracy of AUC verification reached94.25%,thereby constructing agene regulatory network with time-delayed p53 related genes.The results show that the model has a good verification relationship in screening out the gene pairs with demonstrated regulatory relationships,and constructs a visual gene regulatory network.It has good predictive value and research value in screening out gene pairs with unknown regulatory relationship.更多还原