UC San Diego

The role of computation in drug discovery has increased considerably in the past few decades. As of 2004 about 50 compounds discovered with computational approaches had entered human clinical trials and some are FDA approved(1). Computational methods can be employed to effectively guide medicinal chemists and discover and optimize active compounds(2). Here we have focused on two protein targets, which are medically important, to inhibit with novel compounds that we have discovered virtually. In chapter 2 we describe the discovery of the first inhibitors of the Akt phosphatase PHLPP, (PH domain Leucine-rich repeat Protein Phosphatase). This enzyme opposes the effects of two kinases, Akt and PKC, which play a major role in cell growth and survival making it an attractive drug target in the treatment of diabetes and heart disease(3, 4). We have used a combination of in silico and chemical screening tools to find new inhibitor compounds of PHLPP. In chapter 3 we describe drug discovery efforts towards inhibitors of the second target, which is undecaprenyl pyrophosphate synthase, (UPPS) an enzyme necessary for bacterial cell wall synthesis. This recently discovered enzyme is substantially different than the analogous human enzyme making it an attractive target for anti-bacterial drug discovery(5). However UPPS is highly flexible target showing great changes in the active site upon substrate, product, or inhibitor binding, thus making it challenging for structure based drug design. Therefore we have characterized these changes using molecular dynamics simulations, and extracted rare conformations important to drug design. In addition to applying existing methods to new drug targets we have worked on developing a new form of enhanced sampling MD described in chapter 4. This method relies on the same approach as Hamelberg et al. took in accelerated MD simulations but the mathematical equations modifying the potential energy surface increase the recovery of statistics from the simulation. This new boost equation has been applied to statistically rigorous free energy calculations that incorporate solvent, entropy, and dynamics, and are among the most computationally accurate yet costly calculations of ligand-protein binding free energy