Plant & Animal Genomes XIX Conference

January 15-19, 2011
Town & Country Convention Center
San Diego, CA

Prediction Of MicroRNA Targets Using Common Features Of MicroRNA Classes

Bram Sebastian¹ , Samuel E Aggrey^{1, 2}

¹  Institute of Bioinformatics, University of Georgia, Athens, GA 30602, USA

²  Department of Poultry Science, University of Georgia, Athens, GA 30602, USA

MicroRNAs (MiRNA) are small non-coding RNAs which regulate gene expressions by targeting the mRNAs. Targets for miRNA have been identified using both computational and biological methods in both plants and mammals however target prediction is a great challenge in animals compared to plants. We studied the characteristics of human miR-1 and miR-124 to create a miRNA target prediction program based on the common features of the interaction with their targets. The program yielded 78% sensitivity and 98% specificity for miR-1 target prediction and 77% sensitivity and 98% specificity for miR-124 target prediction. To test whether miRNA class grouping is necessary for miRNA target prediction, we used the features of the interaction of miR-1 and miR-124 to their targets to predict miR-16 and miR-15a targets. We obtained 28% sensitivity and 98% specificity for both miR-16 and miR-15a. This indicates that the features of miR-1 and miR-124 target interaction are different than miR-16 and miR-15a and that miR-16 and miR-15a have their own target interaction characteristics. Hence grouping miRNA by classes significantly improves sensitivity and specificity of miRNA target prediction. We also tested the efficacy of two previously developed programs, miRanda and miTarget to predict miR-16 and miR-15a targets. These programs were developed based on generalized features of miRNA and target interactions. Sensitivity of miRanda was 23% for mir-15a and 0% for mir-16, while miTarget had 14% sensitivity for mir-15a and 16% for mir-16. Specificity for miRanda is 99% for miR-15a and 98% for miR-16. We therefore conclude that developing miRNA target programs based on specific features of a particular MiRNA class and its target interactions is much more efficient than using generalized features.