Feasibility of using probabilistic methods to analyse microRNA quantitative data in forensically relevant body fluids: a proof-of-principle study
Feasibility of using probabilistic methods to analyse microRNA quantitative data in forensically relevant body fluids: a proof-of-principle study
复制标题
DOI：
10.1007/s00414-021-02678-w
复制
发表时间：
2021-09
影响因子：
2.1
通讯作者：
Zhilong Li;M. Lv;Duo Peng;Xiao Xiao-Xiao;Zhuangyan Fang;Qian Wang;Huan Tian;L. Zha;Li Wang;Yu Tan;Weibo Liang;Lin Zhang
中科院分区：
文献类型：
--
作者：
Zhilong Li;M. Lv;Duo Peng;Xiao Xiao-Xiao;Zhuangyan Fang;Qian Wang;Huan Tian;L. Zha;Li Wang;Yu Tan;Weibo Liang;Lin Zhang
自microRNA（miRNAs）被引入体液鉴定领域以来，已有多项研究证实其是一种很有前途的体液鉴定标记物。然而，在参考基因的选择和鉴定策略上没有达成共识。在这项研究中，从三个法医学相关的体液数据集中筛选了13个潜在的候选miRNAs，并使用实时定量方法确定了五种体液中12种标记物的表达。两种概率方法，朴素贝叶斯（NB）和偏最小二乘判别分析（PLS-DA），然后应用于预测样品的来源，以确定概率方法是否有助于使用miRNA定量数据进行体液鉴定。同时，采用14种参比品组合验证了不同参比品选择对预测结果的影响。结果表明，在NB模型中，留一法交叉验证（LOOCV）在大多数参考组合中达到了100%的准确率，测试集的预测准确率也达到了100%。在PLS-DA模型中，当使用miR-92 a-3 p作为参考时，前两个组件可以解释约80%的表达方差，并且LOOCV达到100%的准确性。本研究初步证明概率方法在基于miRNA的体液鉴定中具有巨大的潜力，参考物的选择在一定程度上影响预测结果。
Several studies have confirmed that microRNAs (miRNAs) are promising markers for body fluid identification since they were introduced to this field. However, there is no consensus on the choice of reference genes and identification strategies. In this study, 13 potential candidate miRNAs were screened from three forensically relevant body fluid datasets, and the expression of 12 markers in five body fluids was determined using a real-time quantitative method. Two probabilistic approaches, Naive Bayes (NB) and partial least squares discriminant analysis (PLS-DA), were then applied to predict the origin of the samples to determine whether probabilistic methods are helpful in body fluid identification using miRNA quantitative data. Furthermore, 14 reference combinations were used to validate the influence of different reference choices on the predicted results simultaneously. Our results showed that in the NB model, leave-one-out cross-validation (LOOCV) achieved 100% accuracy and the prediction accuracy of the test set was 100% in most reference combinations. In the PLS-DA model, the first two components could interpret about 80% expression variance and LOOCV achieved 100% accuracy when miR-92a-3p was used as the reference. This study preliminarily proved that probabilistic approaches hold huge potential in miRNA-based body fluid identification, and the choice of references influences the prediction results to a certain extent.