Current Research and Scholarly Interests
The central theme of our research is to develop and apply novel theoretical methods to understand the physical properties of biological molecules, such as proteins, nucleic acids, and lipid membranes, and to apply this understanding to design novel synthetic systems, including small molecule therapeutics. In particular, we are interested in the self-assembly properties of biomolecules: for example, how do protein and RNA molecules fold? How do proteins misfold and aggregate and how can we use our understanding of this process to tackle misfolding related diseases, such as Alzheimer's or Huntington's Disease? How can we design or discover novel small molecules to inhibit this process?
As these phenomena are complex, spanning from the molecular to mesoscopic length scales and the nanosecond to millisecond timescales, our research employs a variety of methods, including statistical mechanical analytic models, Markov State Models, and statistical and informatic methods, as well as Monte Carlo, Langevin dynamics, and molecular dynamics computer simulations on workstations and massively parallel supercomputers, superclusters, and large-scale worldwide distributed computing (see http://folding.stanford.edu). Our work also touches closely in parts with applications of Bayesian statistics to statistical mechanics, as well as novel means for computational small molecule (drug) design (such as novel methods for docking and free energy calculation).
For example, we are currently investigating the nature of protein folding and misfolding, relevant for diseases such as Alzheimers and Huntingtons Disease. We have performed simulations of these processes, in all-atom detail on experimentally relevant timescales (milliseconds to seconds), yielding specific predictions of the structural and physical chemical nature of protein aggregation involved in these diseases. These simulation results have then fed into novel computational small molecule drug design methods, yielding novel chemical entities with important and interesting impact.
Since such problems are extremely computationally demanding, we have developed distributed computing projects for protein folding dynamics ("Folding@Home": http://folding.stanford.edu) which has attracted over 4,000,000 PCs since the project's beginning in October 1, 2000 and today is recognized as the most powerful supercomputer/supercluster in the world. Such enormous computational resources have allowed us to simulate unprecedented folding timescales (microseconds to milliseconds) and statistical precision and accuracy (such as very accurate and precise free energy calculations). For more details, please see http://pande.stanford.edu.
Finally, we also have done extensive work in the application of Machine Learning (ML) to Chemistry and Biophysics. We have pioneered traditional and deep learning approaches to cheminformatics and biophysics. In particular, ML methods have played a key role in MSM methods. Moreover, we have been pioneering ML approaches, especially deep learning, for drug design and related areas. Our vision is that we are just scratching the surface of how ML can impact Chemistry and are positioned to be leaders in this burgeoning field.