Improving performance of mammalian microRNA target prediction

Abstract

BackgroundMicroRNAs (miRNAs) are single-stranded non-coding RNAs known to regulate a wide range of cellular processes by silencing the gene expression at the protein and/or mRNA levels. Computational prediction of miRNA targets is essential for elucidating the detailed functions of miRNA. However, the prediction specificity and sensitivity of the existing algorithms are still poor to generate meaningful, workable hypotheses for subsequent experimental testing. Constructing a richer and more reliable training data set and developing an algorithm that properly exploits this data set would be the key to improve the performance current prediction algorithms.ResultsA comprehensive training data set is constructed for mammalian miRNAs with its positive targets obtained from the most up-to-date miRNA target depository called miRecords and its negative targets derived from 20 microarray data. A new algorithm SVMicrO is developed, which assumes a 2-stage structure including a site support vector machine (SVM) followed by a UTR-SVM. SVMicrO makes prediction based on 21 optimal site features and 18 optimal UTR features, selected by training from a comprehensive collection of 113 site and 30 UTR features. Comprehensive evaluation of SVMicrO performance has been carried out on the training data, proteomics data, and immunoprecipitation (IP) pull-down data. Comparisons with some popular algorithms demonstrate consistent improvements in prediction specificity, sensitivity and precision in all tested cases. All the related materials including source code and genome-wide prediction of human targets are available at http://compgenomics.utsa.edu/svmicro.html.ConclusionsA 2-stage SVM based new miRNA target prediction algorithm called SVMicrO is developed. SVMicrO is shown to be able to achieve robust performance. It holds the promise to achieve continuing improvement whenever better training data that contain additional verified or high confidence positive targets and properly selected negative targets are available.

Analysis of the accessibility of CLIP bound sites reveals that nucleation of the miRNA:mRNA pairing occurs preferentially at the 3′-end of the seed match

“….,7) in all human and mouse 39 UTR sequences were computed with RNAplfold ) using a window W = 80 and a maximum pairing distance L = 40, as used in other studies Betel et al 2010;Liu et al 2010;Li et al 2011;Vanicek 2011, 2012).…”

“…Detailed analyses of proteomics, transcriptomics, and CLIP-seq experiments have confirmed that site accessibility is a useful criterion for discriminating between functional and nonfunctional miRNA binding sites (Hausser et al 2009;Wen et al 2011). Because of its proven role in binding-site identification, accessibility has been employed in several miRNA target prediction methods to reduce the false-discovery rate Muckstein et al 2006;Kertesz et al 2007;Long et al 2007;Busch et al 2008;Betel et al 2010;Liu et al 2010;Vanicek 2011, 2012). Some of these methods have implemented an accessibility filter based on the assumption that only 3-4 nucleotides (nt) in the target should be accessible in order to nucleate the binding with the whole miRNA Long et al 2007;Marin and Vanicek 2011).…”

mirMark: a site-level and UTR-level classifier for miRNA target prediction

“…The CTS density (site_density) is computed as number_sites/UTR_length, as done in SVMicrO [18]. Finally, another measure of density is computed by counting the maximum number of CTSs that lie within 100 nt of each other (max_100_sites).…”

“…to existing publically available miRNA target prediction software: SVMicrO [18], TargetScan [4], MiRanda [9], RNAhybrid [12], and PITA [13]. Only publically available software is considered in order to obtain predictions on the mock miRNAs.…”

“…As the number of experimentally verified pairs has increased significantly over the years, statistical or machine learning methods that are data-driven are becoming popular in the area of non-coding RNA classification, including miRNA target prediction [16,17]. An example of the data-driven miRNA target predictor is SVMicrO [18] that combines a larger variety of features than those of rule-based systems with the popular Support Vector Machine (SVM) learning algorithm [19].…”