CVC Seminar Wed Sept 14 1:15-2:30pm ACES 4.304
Speaker: Qiming Yuan
Title: Learning Parameters in F2Dock
Abstract: Choosing parameters for a scoring function plays an important role in protein-protein docking prediction. Some previous approaches only rely on empirical statistics, positive examples or average negative examples. We formulate a learning framework for deriving parameters for scoring function based on observed binding modes. A QP based optimization algorithm is proposed to maximize the distance between correct and incorrect configurations. Experiments are conducted on scoring function in F2Dock but this framework can handle a much richer class of interaction modes (the only constraint is the scoring function should be a weighted summation of multiple energy terms).