Speaker
Prof.
Karl Freed
(James Franck Institute, University of Chicago)
Description
Successful methods for predicting protein structure from the
amino acid sequence have relied upon machine learning
methods and the homology to sequences of proteins with known
structures. These methods fail when homology is low, when
templates are unavailable for large inserts and/or end
portions (InsEnds), and when the proteins become large or
have multiple domains. Thus, fully “ab initio” methods
without homology or machine learning are necessary to
provide the concepts and tools to attack these major
unsolved problems. We develop an ab initio, iterative Monte
Carlo simulated annealing method for sequentially assigning
secondary structure and for prediction the overall protein
structure. This ItFix method provides structures almost
comparable with homology modeling when homology is adequate,
but ItFix fares well for sequences with low homology and for
InsEnds with as many as 45 amino acids and secondary
structure. Predictions are also generated for the folding
sequence.