Computational analysis of regulatory mechanism and interactions of microRNAs

Abstract

For years, RNAs were thought to have only two broad functions in cells, transmitting information between DNA and protein as messenger RNA (mRNA), and playing structural, catalytic, information decoding roles in protein synthesis as ribosomal RNA (rRNA) and transfer RNA (tRNA). However, the discovery of RNA interference (RNAi) changed this picture. RNAi is a regulatory process that uses small non-coding RNAs (ncRNAs) to suppress gene expression at the post-transcriptional level. This discovery led to identification of many classes of functional ncRNAs. MicroRNA (miRNA) is a class of such ncRNAs with ∼22 nucleotides that are quite abundant and found in most eukaryotic cells. This thesis focuses on revealing regulatory roles and characteristics of miRNAs through bioinformatics approaches by addressing three research questions. The first research question is whether we can enhance miRNAs target prediction in animals by considering multiple target sites. Many algorithms exist for miRNA target predictions, but most algorithms do not consider multiple target sites. Predicting accurate miRNA target genes is important to infer miRNA regulatory roles since annotations of miRNA regulations are poor. To solve this possible fault, we developed a two step support vector machine (SVM) model. Benchmark tests showed that our two step model outperformed other existing miRNA target prediction algorithms. The second research question is whether there are factors to explain differences between different miRNA high-throughput experiments. There are several high-throughput technologies widely used for miRNA experiments, such as microarray and quantitative proteomics, but the results from these technologies are often inconsistent. By statistically analyzing several such high-throughput miRNA experiments, we revealed the characteristic of different technologies and also identified several factors that cause the differences. The third research question is whether miRNAs interact with other classes of ncRNAs. There are strong evidences that some miRNAs are involved in transcription by interacting with other ncRNAs. We investigated ncRNAs in complex loci to find potential miRNA:ncRNA interactions. A complex locus is a locus that contains multiple genes that interact between themselves. We found evidence that some miRNAs are involved in transcriptional regulation with ncRNAs in complex loci. In summary, this thesis provides solutions for these research questions, and it contributes to a better understanding of several important aspects of miRNA characteristics and regulations. It also shows effective bioinformatics approaches to develop a robust machine learning model and analyze different miRNA high-throughput experiments.