Danielle Talbot

Supervisor: Dr Andrew C R Martin

School: School of Animal and Microbial Sciences, University of Reading

Generating the correct alignment between template and target is the most important stage of comparative modelling. When the alignment is incorrect it will introduce major errors into the final modelled structure.

Simple sequence alignments do not always give the correct structural aligment, especially when the sequence identity between the template and the target is low. For example, indels within regions of secondary structure are relatively rare - a pure sequence alignment will not reflect this and may place an indel within a section of secondary structure rather than shifting it to an adjacent loop.

One way to try to improve comparative modelling is to create many models each with a different alignment. The resulting models are then analysed and scored to select the model likely to be most accurate. We thus need to address two problems: (1) how to generate different possible alignments (2) how to score those alignments.

To address the first problem we have started by examining in detail the extreme case of alignment error known as MLSAs (Misleading Local Sequence Alignments). In these relatively rare cases, the target and template sequence have an apparently obvious local alignment with high sequence identity, but the structural alignment between the two proves very different. We have identified a number of factors which appear to contribute to these mis-alignments.

For the second problem, we have investigated the application of empirical potentials to distinguish between models created from different alignments. The RAM potential set was used to attempt to distinguish between models created from a sequence alignment and the (correct) structural alignment using approximately 50,000 template/target pairs. In the majority of cases the model derived from the structural alignment scored better than that derived from the sequence alignment. However, when applied to distinguishing between models with more subtle variations in alignment, or simply different loop conformations, the RAM potential did not perform very well.