Figure 1

Protein alignment as a Low Order Value Optimization problem. (a) For each correspondence between Cα atoms of proteins A and B, there is a smooth score that depends on the rotations and translations of the proteins. (b) In protein alignment, the objective function is the function that assumes the maximum value among all the possible score functions. (c) An algorithm that converges to maximizers must have two main steps: A step (A) that recognizes the best correspondence and a step (B) that maximizes the score given the correspondence. Note that at point C the objective function is non-smooth.