Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Biochemistry

Supervisor

Dr. Gregory B. Gloor

Abstract

A major goal in computational biochemistry is to obtain three-dimensional structure information from protein sequence. Coevolution represents a biological mechanism through which structural information can be obtained from a family of protein sequences. Evolutionary relationships within a family of protein sequences are revealed through sequence alignment. Statistical analyses of these sequence alignments reveals positions in the protein family that covary, and thus appear to be dependent on one another throughout the evolution of the protein family. These covarying positions are inferred to be coevolving via one of two biological mechanisms, both of which imply that coevolution is facilitated by inter-residue contact. Thus, high-quality multiple sequence alignments and robust coevolution-inferring statistics can produce structural information from sequence alone. This work characterizes the relationship between coevolution statistics and sequence alignments and highlights the implicit assumptions and caveats associated with coevolutionary inference. An investigation of sequence alignment quality and coevolutionary-inference methods revealed that such methods are very sensitive to the systematic misalignments discovered in public databases. However, repairing the misalignments in such alignments restores the predictive power of coevolution statistics. To overcome the sensitivity to misalignments, two novel coevolution-inferring statistics were developed that show increased contact prediction accuracy, especially in alignments that contain misalignments. These new statistics were developed into a suite of coevolution tools, the MIpToolset. Because systematic misalignments produce a distinctive pattern when analyzed by coevolution-inferring statistics, a new method for detecting systematic misalignments was created to exploit this phenomenon. This new method called ``local covariation'' was used to analyze publicly-available multiple sequence alignment databases. Local covariation detected putative misalignments in a database designed to benchmark sequence alignment software accuracy. Local covariation was incorporated into a new software tool, LoCo, which displays regions of potential misalignment during alignment editing assists in their correction. This work represents advances in multiple sequence alignment creation and coevolutionary inference.


Share

COinS