Prediction of protein binding and folding from molecular geometry and empirical scoring functions

Two fundamental processes of living cell, protein binding and protein folding, were studied through building computational models on experimental data. A protein-protein docking program and a protein folding rate prediction tool were developed.;We presented a novel geometry-based docking algorithm with effective conformation-search and improved scoring functions for predicting the structure of protein-protein complex. Ligand proteins docking only need undergo transformation and rotation along surface normal vectors. The computational time of our algorithm is approximately proportional to O(mxn). The hierarchical scoring scheme improved the scoring performance by clustering protein binding complexes according to the energy score distribution of sampled interfaces. A total of 330 protein complexes were collected to build and evaluate our method. To predict protein folding rate from amino acid sequence, we introduced two new quantity, Nalpha, which is a count of the number of well-packed nonlocal contacts in the native structure of a protein, and NNP (nearest neighbor pair) which is a pair of residues which are nearest neighbor of each other in the primary sequence.;Compared to existing interface prediction methods, our method shows higher success rates: 71.4% for enzyme-inhibitors and 44.0% for other interfaces. Compared with other docking methods, we achieved similar level of performance: 60% and 40% success rate within top 40 predictions respectively. For protein folding rate prediction, our method shows an excellent correlation with experimental observations; the correlation coefficients are 0.98 for all tested proteins. These accuracy levels are superior to any other methods in the literature. And our method is the first that solely uses the amino acid sequence information alone for predicting the folding rates of proteins.