|
Abstract:
|
The problem of predicting a protein's 3-D structure from its amino acid sequence, using computational methods, has captured the attention of scientists for the last half century. The structure of a protein is essential for understanding its function, and hence accurate structure prediction is of vital importance in modern applications such as protein design in biomedicine. In this talk, I will highlight a specific application: design of the H3 loop of an antibody and prediction of the resulting antibody-antigen complex. An effective general strategy for this problem is first sampling candidate conformations, then ranking the candidates, for example according to an energy function. However, the size of the conformational space poses a significant computational challenge. Hence we first adapt a recently developed sequential Monte Carlo algorithm, to tackle efficient sampling of low-energy conformations for H3 loops. We then build a ranking model from the samples, to attain higher prediction accuracies than using an energy function alone. Our methods were tested in a blinded prediction challenge and promising results were obtained.
|
