Author's School

Graduate School of Arts & Sciences

Author's Department/Program

Biology and Biomedical Sciences: Computational and Systems Biology


English (en)

Date of Award

January 2009

Degree Type


Degree Name

Doctor of Philosophy (PhD)

Chair and Committee

Sean Eddy


Functional RNA elements do not encode proteins, but rather function directly as RNAs. Many different types of RNAs play important roles in a wide range of cellular processes, including protein synthesis, gene regulation, protein transport, splicing, and more. Because important sequence and structural features tend to be evolutionarily conserved, one way to learn about functional RNAs is through comparative sequence analysis - by collecting and aligning examples of homologous RNAs and comparing them. Covariance models: CMs) are powerful computational tools for homology search and alignment that score both the conserved sequence and secondary structure of an RNA family. However, due to the high computational complexity of their search and alignment algorithms, searches against large databases and alignment of large RNAs like small subunit ribosomal RNA: SSU rRNA) are prohibitively slow. Large-scale alignment of SSU rRNA is of particular utility for environmental survey studies of microbial diversity which often use the rRNA as a phylogenetic marker of microorganisms. In this work, we improve CM methods by making them faster and more sensitive to remote homology. To accelerate searches, we introduce a query-dependent banding: QDB) technique that makes scoring sequences more efficient by restricting the possible lengths of structural elements based on their probability given the model. We combine QDB with a complementary filtering method that quickly prunes away database subsequences deemed unlikely to receive high CM scores based on sequence conservation alone. To increase search sensitivity, we apply two model parameterization strategies from protein homology search tools to CMs. As judged by our benchmark, these combined approaches yield about a 250-fold speedup and significant increase in search sensitivity compared with previous implementations. To accelerate alignment, we apply a method that uses a fast sequence-based alignment of a target sequence to determine constraints for the more expensive CM sequence- and structure-based alignment. This technique reduces the time required to align one SSU rRNA sequence from about 15 minutes to 1 second with a negligible effect on alignment accuracy. Collectively, these improvements make CMs more powerful and practical tools for RNA homology search and alignment.


Permanent URL: