Computational modeling of splicing regulation

Abstract

Alternative splicing is one of the most important post-transcriptional modification in cell. It increases the coding capacity of the genome by enable one gene encoding multiple proteins. The majority of human protein-coding genes undergo alternative splicing. And mis-splicing of those genes are known to be associated with many human diseases. Therefore, it is important to study and understand the splicing regulatory machinery. The splicing regulation consists of two components: transacting regulators and cis-acting elements. In this dissertation, we explored these two aspects of splicing regulation. First, we investigate the relationship of three key trans-acting regulators: hnRNP A1, SRSF1 and U2AF with transcriptome-wide individual-nucleotide resolution cross-linking and immunoprecipitation (iCLIP) data. Our result revealed the competition relationship between hnRNP A1 and SRSF1 on 3’ splicing sites, and the inhabitation effects on U2AF recruitment after hnRNP A1 overexpression. We also discovered that Alu elements may serve as cis-acting elements and compete with authentic exons for the binding of U2AF. Second, we developed a machine learning algorithm to prioritize the disease-causing probability of intronic single-nucleotide variants (iSNVs) by evaluating their cisacting impact on both alternative splicing and protein structure. The resulting predictive model can predict pathogenic iSNVs with high accuracy and outperform popular algorithms such as splicing-based analysis of variants (SPANR) and combined annotation–dependent depletion (CADD). This suggests that protein structure features can provide additional layer of information in prioritizing pathogenic iSNVs. In conclusion, our studies provide remarkable insights on alternative splicing regarding both trans-acting regulation and cis-acting regulation. The discoveries of our research on trans-acting regulators are valuable for understanding splicing regulatory machinery. The algorithm we developed can be used to prioritize pathogenic iSNVs without needing to test them all in expensive and laborious assays.