Tuesday,
October 20, 12:00 p.m. ECG 317
Mayo Clinic, Scottsdale and Biomedical Informatics, ASU
Computational Immunology:
Prediction of Peptides involved in the T-cell
Immune Response
Abstract
We will discuss two complementary computational methods for predicting
which peptides bind to class II MHC proteins. These immune system
proteins bind fragments of extracellular proteins from pathogens and
thereby activate a T cell immune response. Knowledge of which fragments
bind to a particular MHC can be used to design vaccines and better
understand immune system function and dysfunction, such as autoimmune
diseases. Computational methods are needed because the large number of
different MHC proteins and possible peptide sequences preclude
comprehensive experimental testing. We have developed both
sequence-based methods as well as structure-based methods that use
available experimental protein structures to predict binding
affinities. The accuracies of these methods were evaluated by
comparison with experimental peptide binding and structure data. The
sequence-based method, which incorporates thermodynamic averaging and
regularization, was found to perform favorably compared with other
prediction methods. Preliminary results were also obtained for the
structure-based method. Although slower, it was able to make accurate
predictions for a different MHC type than that used for training the
method. This is an important goal as experimental training data is only
available for a small fraction of MHC types. Finally, we will describe
future extensions of these prediction methods.
Part of this is joint work with Hans Mittelmann, School of Math&Stats, ASU
