Abstract 55: Prediction of miRNA target genes using a hidden Markov model

Abstract

Abstract MicroRNAs (miRNAs) are short non-coding RNAs about 22 nucleotides in length that play important regulatory roles in animals for translational repression by targeting mRNAs. Recent studies reported that miRNAs may function as tumor suppressors or oncogenes, and the alterations in miRNA expression may be critical in tumor genesis and cancer progression. Using computational methodologies can rapidly identify miRNA targets in massive data, and provide rich gene-related information for human cancer studies. However, it is challenging to predict targets in animals due to imperfect complementarity between miRNAs and mRNA targets. In order to further improve the prediction performance, we proposed a novel miRNA target-gene prediction algorithm that combines several conventional prediction models including the sequence complementary searching for calculating alignment scores and thermodynamic stability approaches for assigning folding free energy to each miRNA-target interaction. A Hidden Markov Model (HMM), a well known machine learning approach, was implemented in the algorithm to help the prediction decision. However, an HMM cannot consider all global information of the sequences due to its innate limitation. Therefore, forward and backward HMMs were simultaneously utilized in the proposed algorithm to overcome this limitation. As a result, any element information of miRNA-target interactions was able to pass to any other element by bi-directions. The proposed algorithm can predict the cancer-related miRNA target genes through a suitable combination of the proposed models. The results show that the highest sensitivity, specificity, and overall accuracy were 84.25%, 96.78%, and 96.67%, respectively. Based on our analysis, 52.42% prediction result of the proposed algorithm was consistent with other existing prediction algorithms. Furthermore, the predicted target genes obtained by the proposed algorithm could be likely found from the potential cancer-related genes discovered by microarray data. A total of 70.37% predicted target genes were significantly overlapped with the cancer-related genes from microarray data, suggesting these genes were substantially regulated by miRNAs in cancer. In conclusion, our proposed novel algorithm is able to predict miRNA target genes, commonly found among microarray-identified dysregulated genes in previous cancer studies. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 55. doi:10.1158/1538-7445.AM2011-55