基于上下游转录调控的特征数据识别关键microRNA

Identification of key microRNAs based on feature of upstream and downstream transcriptional regulation

近年来,基于机器学习的microRNA与复杂疾病之间的关系预测受到广泛关注。然而现存的方法大多是围绕microRNA表达谱,相似性网络,序列特征等角度开发的,忽略了转录调控信息的重要性。基于此提出了一种新的基于microRNA上下游转录调控特征数据的机器学习算法(RPDZ)去预测癌症相关microRNA的重要性。此方法首先通过手动搜索和批量处理高通量数据,计算和整合microRNA上下游的特征数据,然后利用随机森林进行有效的特征提取,最后利用深度神经网络更好地捕捉疾病与microRNA的关系,更好地预测microRNA的重要性得分。使用ROC曲线下方的面积(AUC)、F-measure值和准确度(ACC)作为评估预测性能的指标。通过五折交叉验证的实验表明所提出的算法(ACC:0.874 8,AUC:0.93,F-measure:0.870 1)相比其他三种对比方法PESM、SVM、GaussianNB可得到更好的分类识别效果。算法不仅可以有效地整合多组学特征数据识别关键的miRNA,还可以为精确和自动化的计算机辅助诊断奠定基础。

In recent years,the prediction of the relationship between microRNAs and complex diseases based on machine learning has received widespread attention.However,most of the existing methods focus on microRNA expression profiles,similarity networks,sequence characteristics and so on,ignoring the importance of transcriptional regulatory information.Based on this,a novel microRNA upstream and downstream transcriptional regulation feature based machine learning algorithm(RPDZ) was proposed to predict the importance of cancer-related microRNA.First,high-throughput data is searched and proposed,the feature of microRNA upstream and downstream is calculated and integrated,then random forest is used for effective feature extraction,and finally the disease-mirna relationships is captured and the importance of microRNA score is predicted with the deep neural network.The area under the ROC curve(AUC),F-measure and accuracy(ACC) as indicators is used to evaluate the prediction performance.The five-fold cross-validation experiment shows that our proposed algorithm can obtain better classification and recognition results(ACC:0.8748,AUC:0.93,F-measure:0.8701)than the other three comparison methods,PESM,SVM and GaussianNB.This algorithm can not only effectively integrate multiple omics feature data to identify key miRNAs,but also lay the foundation for accurate and automated computer-aided diagnosis.