In protein structure prediction, it is a quite common task to search for energetically feasible protein loop conformations connecting parts of the protein of higher structural order (secondary structure). The problem of predicting native-like loop structures is widely known as the protein loop closure problem. Recently published results are an evidence for the progress in the accuracy of the predictions and the efficiency of loop closure methods in general.
Protein loop closure is formulated as the search for loop structures in conformation space that satisfy the loop closure constraints, i.e. structures that fill the gap between known secondary structures of a protein fixed in space. The focus of this thesis is to improve the conformational sampling for closed loop conformations. A completely novel approach to this problem inspired from robotics using a mechanistic description of the loop chain and a motion planning technique is proposed.
There are indications that enhancements of the conformational sampling methodology improve the accuracy of protein structure predictions. It is supposed that the loop modeling algorithm presented in this thesis will potentially yield loop conformations of lower energy and closer to the native structure when compared to other methods. The thesis serves as a proof-of-concept study for the general applicability of the presented method to protein loop modeling.
The mechanistic description of the protein loop structure is based on the representation by a kinematic chain inspired from robot modeling. The transpose of the Jacobian matrix computed from the kinematic chain representation relates generalized forces acting on the amino acid residue at the free end of the loop chain (end-effector) to torques around the torsion angles of the protein backbone of the loop chain. Self-motions of the kinematic chain due to its redundancy in the number of degrees-of-freedom (DOF) are used to minimize an energy function. An iterated motion scheme is derived from this mechanistic description.
The motion scheme is used as a local planner for a randomized motion planning technique based on Rapidly-exploring Random Trees (RRT). The motion planning algorithm developed in this thesis combines aspects from task space and configuration space planning. It is incorporated into the loop modeling application of the Rosetta protein modeling suite.