Abstract
I am going to talk about a problem I encountered during my rotation with Dr. Hal Stern. We analyzed data from a project in which Dr. Kai-Ming Ho's group is developing a fast protein structural "threading" method. Threading is an approach for predicting the three-dimensional structure of a protein by aligning the amino acid sequence of a "target" protein with representative "template" protein structures. A scoring function is used to evaluate target-template alignments. The best-scoring alignment should identify the template structure most compatible with the target sequence and thus "predict" a meaningful structural model for the target protein. In order to assess the significance of each alignment score (say, X0), the target amino acid sequence is randomly permuted and aligned with the same template structure to get a score Xi ( i = 0, 1, ...). This process is repeated until a decision is reached, with Pr ((X0 - mu) > 6 | X1, ..., Xn) or Pr((X0 - mu) / sigma > 2 | X1, ..., Xn) as the decision criteria. The question we addressed is how to compute these two probabilities. We have tried to approach this question in several ways, e.g., Bayesian framework, non-parametric methods. I will discuss these approaches and present some of the results obtained.