人源microRNA前体的全基因组预测

microRNA（miRNA）是一类不编码蛋白的调控小分子RNA,在真核生物中发挥着广泛而重要的调控功能.由于miRNA的表达具有时空特异性,因而通过计算方法预测miRNA而后有针对性的实验验证是miRNA发现的一条重要途径.降低假阳性率是miRNA预测方法面临的重要挑战.本研究采用集成学习方法构建预测miRNA前体的分类器SVMbagging,对训练集、测试集和独立测试集的结果表明,本研究的方法性能稳健、假阳性率低,具有很好的泛化能力,尤其是当阈值取0.9时,特异性高达99.90%,敏感性在26%以上,适合于全基因组预测.采用SVMbagging在人全基因组中预测miRNA前体,当取阈值0.9时,得到14933个可能的miRNA前体.通过与高通量小RNA测序数据的比较,发现其中4481个miRNA前体具有完全匹配的小RNA序列,与理论估计的真阳性数值非常接近.最后,对32个可能的miRNA进行实验验证,确定其中2条为真实的miRNA.

Genome-wide Prediction of Human microRNA Precursors

microRNAs (miRNAs) are a class of small regulatory non-coding RNAs. They are involved in diverse pathways and play important roles in gene regulation in eukaryotes. Since the expression of miRNAs is spatial and temporal-specific, computational prediction followed by experimental validation is still an important approach in miRNA discovery. Decreasing false positive ratio is very challenging in miRNA prediction. Here we employed ensemble learning to construct the classifier SVMbagging to predict miRNA precursors. The results of the training, test and independent test datasets demonstrate that SVMbagging is robust, has low false positive ratio and good generalization ability. Especially when the threshold is 0.9, the specificity of SVMbagging is as high as 99.90% and the sensitivity is higher than 26.00%. Therefore, SVMbagging is suitable for genome-wide prediction. We applied SVMbagging in genome-wide prediction of miRNA precursors in human genome. We obtained 14933 candidate miRNA precursors at the threshold of 0.9. Among them, 4481 miRNA precursors have perfect matches with small RNA reads when aligning with three groups of small RNA datasets from high-throughput sequencing technologies. The number of miRNAs is very close to the true positives estimated theoretically according to the performance of SVMbagging. Finally, we applied experimental methods to validate 32 candidate miRNAs. Two of them were confirmed to be true miRNAs.