systems MOIL and LOOPP available on the web http://www.tc.cornell.edu/reports/NIH/resource/CompBiologyTools/.
Current research directions include:
Mean field approaches for global optimization and structure prediction (Locally Enhanced Sampling): Structures are often determined by an optimization of an energy function. I introduced mean field approaches that modify the target function and make it more accessible to global optimization. We have applied these techniques to determine conformations of short peptides and to refine low-resolution structures of proteins. These approaches are implemented into MOIL.
Development of folding potentials using linear programming: The design of folding potentials relies on considerable human intuition and many trials and errors. I developed an automated protocol that “learns” from experience and failures and constantly improves the quality of the current potential energy. We prove that the widely used pairwise interaction model cannot recognize exactly correct protein folds. We specifically design energy functions for which threading and folding are performed efficiently and accurately. Based on these studies, an efficient and accurate threading algorithm to recognize protein function was designed. The algorithm fits a sequence to a structure and was implemented in the program LOOPP. In a recent publication in Science we suggested an evolutionary link between a gene that controls the size of the tomato fruit and a protein that participates in controlling cell growth and cancer (joint work with Steve Tanksley’s group).
Extending the time scale of simulations. One of the striking observations in dynamics of biological molecules is the extremely large time scale they covered. Initiation by light absorption of biochemical processes is very rapid (femtoseconds), while protein folding is slow (milliseconds to minutes). Current simulation approaches (Molecular Dynamics MD) are restricted to nanoseconds (10-9 seconds). I developed a stochastic path integral formulation that provides a numerically stable trajectory for almost an arbitrary time step. We apply the new algorithm to study activation of proteins (the R->T transitions in hemoglobin, microseconds) and to protein folding (folding of C peptide, tens of nanoseconds). The method provides systematic approximation to the dynamics and is more efficient than MD by orders of magnitude. It is available in MOIL.
Acting head: NIH resource for parallel computing at the Cornell Theory Center.
Head: Computational Genomics Committee.
Committees: Genomics Curriculum; Graduate Admissions; Computational Genomics Committee for the collaborative efforts at Cornell, Rockefeller, and Sloan Kettering Institutes.
The R->T transition in hemoglobin.
Workshop on Protein Dynamics, Telluride, CO, July 1999.