Date of Award
Doctor of Philosophy (PhD)
Genetics and Biochemistry
Dr. Liangjiang Wang, Committee Chair
Dr. Weiguo Cao
Dr. Charles Schwartz
Dr. Michael Sehorn
The development of bioinformatics methods for human genetic studies utilizes the vast amount of data to generate new valuable information. Machine learning and statistical coupling analysis can be used in the study of human diseases. These diseases include intellectual disabilities (ID), prevalent in 1-3% of the population and caused primarily by genetics. Although many cases of ID are caused by mutations in protein-coding genes, the possible involvement of long non-coding RNAs (lncRNAs) in ID due to their role in gene expression regulation, has been explored. In this study, we used machine learning to develop a new expression-based model trained using ID genes encoded with the developing brain transcriptome. The model was fine-tuned using the class-balancing approach of synthetic over-sampling of the minority class, resulting in improved performance. We used the model to predict candidate ID-associated lncRNAs. Our model identified several candidates that overlapped with previously reported ID-associated lncRNAs, enriched with neurodevelopmental functions, and highly expressed in brain tissues. Machine learning was also used to predict protein stability changes caused by missense mutations, which can lead to disease conditions including ID. We tested Random Forests, Support Vector Machines (SVM) and Naïve Bayes to find the best-performing algorithm to develop a multi-class classifier. We developed an SVM model using relevant physico-chemical features after feature selection. Our work identified new features for predicting the effect of amino acid substitutions on protein stability and a well-performing multi-class classifier solely based on sequence information. Statistical approaches were used to analyze the association between mutations and phenotypes. In this study, we used statistical coupling analysis (SCA) to cluster disease-causing mutations and ID phenotypes. Using SCA we identified groups of co-evolving residues, known as protein sectors, in ID protein families. Within each distinct sector, mutations associated with different phenotypic manifestations associated with a syndromic ID were identified. Our results suggest that protein sector analysis can be used to associate mutations with phenotypic manifestations in human diseases. The bioinformatic methods developed in this dissertation can be used in human genetic research to understand the role of new genes and proteins in human disease.
Coto, Jose Andres Guevara, "Development of New Bioinformatic Approaches for Human Genetic Studies" (2017). All Dissertations. 2039.