Data analysis and bioinformatics
PReMI'07 Proceedings of the 2nd international conference on Pattern recognition and machine intelligence
| Hi-index | 0.01 |
The process of gene splicing, which involves the excision of introns from a pre-mRNA and joining of exons into mature mRNA is one of the essential steps in protein production. Although this process has been extensively studied, it is still not clear how the splice sites are accurately identified and correctly paired across the intron. It is currently believed that identification is accomplished through base-pairing interactions between the splice sites and the spliceosomal snRNAs. However, the relatively conserved sequences at the splice sites are often indistinguishable from similar sequences that are not involved in splicing. This suggests that not only sequence but other features of pre-mRNA may play a role in splicing. A number of authors have studied the effects of pre-mRNA secondary structure on splicing, but these studies are usually limited to one or a small number of genes, and therefore the conclusions are usually gene-specific. This thesis aims to complement previous studies of the role of pre-mRNA secondary structure in splicing by performing a comprehensive computational study of structural characteristics of Saccharomyces cerevisiae introns and their possible role in pre-mRNA splicing. We identify long-range interactions in the secondary structures of all long introns that effectively shorten the distance between the donor site and the branchpoint sequence. The shortened distances are distributed similarly to the branchpoint distances in short yeast introns, which are presumed to be optimal for splicing, and very different from the corresponding distances in random and exonic sequences. We show that in the majority of cases, these stems are conserved among closely related yeast species. Furthermore, we formulate a model of structural requirements for efficient splicing of yeast introns that explains previous splicing studies of the RP51B intron. We also test our model by laboratory experiments, which verify our computational predictions. Finally, we use different computational approaches to identify any structural context at the boundaries or within yeast introns. Our study reveals statistically significant biases, which we use to train machine learning classifiers to distinguish between real and pseudo splice sites.