HOME Protein Structure Prediction MUFOLD-DB Clustering Members
Introduction
We have developed MUFOLD, a hybrid methodology of using whole and partial template information and along with new computational techniques for protein tertiary structure prediction.

Rather than the existing optimization of physical forces to build structural models for given protein sequence, We developed a new distance matrix sampling methodology to achieve more efficient and accurate modeling. On one hand, the detected globally conserved regions (as mention above) represent the structure topology (or skeleton) of the target, thus, the corresponding spatial constraints should be satisfied with little changes to keep the topology undistorted during modeling process. On the other hand, the restraints retrieved from the remainder non-conserved region can guide a deep conformation search space. Specifically, a sample, i.e., the pairwise (including the conserved and un-conserved) spatial constraints among residues, composes a distance matrix. Upon this distance matrix, a MDS (multidimensional scaling) technique was applied to calculate the location of each residue, thus the structural model can be build by the locations of residues. By this way, the modeling process was optimized by best utilizing most relevant information from the templates in PDB.