Fast rank-based normalization of miRNA qPCR arrays using support vector regression

Abstract

One of the first important steps in elucidating the function of microRNAs (miRNAs) is expression profiling. Many methods including low-density qPCR arrays are increasingly used to profile the expression of miRNAs. Normalization techniques are necessary due to certain biases in profiling approaches, and the techniques can significantly affect the accuracy of miRNA quantification. Most normalization methods for continous expression data have been developed for mRNA microarrays and new and modified methods should be used for miRNA studies in general and RT-qPCR miRNA arrays in particular. Previously, cyclic normalization using support vector regression has been successfully applied to mRNA arrays. Here, a new method based on support vector regression is introduced for miRNA normalization and the cyclic nature of algorithm in cyclic spline normalization has also been modified. It was shown that by creating a baseline array, it is possible to remove the cyclic nature of the normalization to achieve faster normalization, with no loss of accuracy. To assess how much the mentioned normalization method reduces technical error, mean square error (MSE) in two real miRNA qPCR array datasets and a simulated dataset before and after normalizations was robustly modelled and compared. Our method was also systematically compared with the most commonly used methods for normalization of qPCR miRNA arrays. The new method showed lower MSE values corresponding to other common methods of miRNA normalization.

Keywords

• microRNA (miRNA) profiling
• Gene expression
• Normalization
• Support vector regression
• Mean square error