A major bottleneck for studying biomolecular systems is the accuracy and availability of consistent molecular mechanical (MM) models, which in turn depend on accurate representation of the potential energies (referred to as the “force field (FF)”). A high-quality force field is the key to successfully model the structures, energies and dynamics of biomolecules as well as to successfully describe biological processes on many levels (folding, binding, cell signaling, etc.) using molecular dynamics (MD) simulations. We have developed a set of popular AMBER force fields that have a great impact in the fields, such as FF99, GAFF (General AMBER Force Field) and PTMFF. Molecular mechanical toolkits (such as Antechamber) have also been developed to facilitate our users to generate molecular mechanical models for arbitary molecules. Now our focus is on the development of the next generation AMBER force field which explicitly includes the energy term of polarization energy calculated using Thole's atomic dipol interaction models. The new polarizable AMBER force field enables us to calculate the energies much more accurately.

In silico modeling of phenotypes and clinical outcomes using pharmcometric and system biology approaches has a great application in drug discovery, regulatory decision and rational drug treatment in patients. We are particually interested in the following two research fields: (1) predictive gene signature using gene network analysis, and (2) new algorithms to improve the successful rates of clinical outcome prediction.

Computer-aided drug design is an indispensable process in modern drug discovery. The CADD methods can be roughly classified into two categories based on whether biologic targets are used in the methods or not. In most cases, a drug is an organic small molecule that activates or inhibits the function of a biomolecule (such as a protein), which in turn results in a therapeutic benefit to the patients. In the most basic sense, drug design involves the design of small molecules that are complementary in shape and charge to the biomolecular target with which they interact and therefore will bind to it. We research interests CADD including drug lead identification by virtual screening, novel docking scoring function development and in silico modeling of ADME-Tox properties.

Atomistic simulations of biomolecules provide a detailed view of structure and dynamics that complement experiments. Increased conformational sampling, enabled by new algorithms and growth in computer power,20 now allows a much broader range of events to be observed, providing critical insights largely inaccessible to experiments. In the last few years, the successful application of graphics process units (GPUs) to MD simulations has dramatically extended our ability to simulate very big systems (>100,000 atoms) at a relative longtime scale (microseconds).21, 22 AMBER is a leading package in the arena of accelerating MD simulations with GPUs. Our recent progress on crystal simulations has greatly benefited from the GPU acceleration.

More and more evidence supports the concept that the amino acid residues engaged in functionally important interactions tend to be evolutionarily conserved, i.e. the involved protein sites are coupled. This concept has been verified experimentally for many protein systems. Unfortunately, most physical scoring function-based protein design algorithms including Rosetta18 lack a term to account for the long-range site-site couplings. This may explain why it is so challenging to engineer dynamic control into proteins and design protein-protein interfaces with those physics-based approaches. In contrast, the sequence-based protein design algorithms represented by statistical coupling analysis (SCA)15 take the long-range site-site couplings into account but cannot filter out those sequences with energetic flaws. We are working on new algorithms to overcome the challenges in protein design by incorporating the site-site correlation information into a scoring function-based protein design scheme.