Current Research Projects

 

Currently we work in the two major areas described below.  There are several common threads that unite these projects.  First, the underlying approach for each is to employ physics-based energy models (all-atom force fields and implicit solvent) for predictive protein modeling.  Second, each project involves a combination of algorithmic development and biological applications; the exact balance between the two is largely the choice of the student working on the project.  Collaborations are highly encouraged.  Third, many of the projects represent areas of computational biochemistry in which little work has been done or little progress has been made. 

 

1.  Atomic-Level Mechanisms of Protein Regulation

Post-translational protein phosphorylation

The goal of this project is to better understand and predict how phosphorylation modulates the energy landscapes of proteins, driving changes in conformation and dynamics, which in turn can modulate binding or activity.  Funding for this project is provided primarily by an NSF CAREER award.  Some key publications:

Groban, Narayanan, and Jacobson. "Phosphorylation-Induced Conformational Changes in Protein Loops and Helices", PLOS Computational Biology, 2 (2006) 238. Online

Mandell, Chorny, Groban, Wong, Levine, Rapp, and Jacobson. "The strengths of hydrogen bonds involving phosphorylated amino acid side chains", JACS, 129 (2007) 820. Online

Our future work in this area will increasingly focus on making testable predictions about mechanisms of phospho-regulation, in collaboration with experimental scientists at UCSF and elsewhere.  

pH regulation

In close collaboration with Diane Barber (cell biology) and Mark Kelly (NMR), we are investigating how pH regulates the function of specific proteins and entire networks.  Our initial focus is on proteins involved in regulation of actin-based motility. At the atomic level, pH regulation and phosphorylation both act as electrostatic switches.  I’m particularly intrigued by how small changes in cytosolic pH can lead to large changes in protein binding or activity. 

Allostery

This is a relatively new project in collaboration with Jim Wells (UCSF), and carried out by our joint student Chris McClendon.  We are using a variety of atomistic simulation methods to tease out how information about binding at one site is transmitted to another site. 

 

2.  Protein-Ligand Interactions and Computer-Aided Drug Design

Funding for these projects is currently provided by 3 NIH program project grants in which I participate as a co-investigator. 

Physics-based scoring of protein-ligand interactions

We have demonstrated that it is possible to use molecular mechanics energy models with implicit solvent (Generalized Born) in high-throughput docking (10s to 100s of thousands of ligands, at a computational expense of less than 1 minute per ligand on a single processor), and that doing so can improve the quality of the results relative to standard docking scoring functions.  We have developed and tested this approach in a series of articles:

Kalyanaraman, Bernacki, and Jacobson. "Virtual screening against highly charged active sites: Identifying substrates of alpha-beta barrel enzymes", Biochem. 44 (2005) 2059. Online

Bernacki, Kalyanaraman, and Jacobson. "Virtual Ligand Screening Against E. Coli Dihydrofolate Reductase: Improving Docking Enrichment Using Physics-Based Methods", J. Biomol. Screening, 10 (2005) 675. Online

Huang, Kalyanaraman, Irwin, and Jacobson. "A Physics-Based Scoring Method Improves Early Enrichment in Virtual Screening of Large Compound databases", J. Chem. Info.  Model. 46 (2006) 243. Online

Huang, Kalyanaraman, Bernacki, and Jacobson. "Molecular mechanics methods for predicting protein-ligand binding", Phys. Chem. Chem. Phys. 8 (2006) 5166. Online

Docking against flexible receptors and homology models

Structure-based drug design is always challenging but particularly so when trying to target a flexible binding site or a protein that lacks any experimental structure.  We have contributed to new methods for these challenges, and we have applied these methods to inhibitor discovery and functional annotation.  Our publications in this area include (several more on the way):

Kenyon, Chorny, Holman, and Jacobson. "Novel human lipoxygenase inhibitors discovered using virtual screening with homology models", J. Med. Chem., 49 (2006) 1356. Online

Sherman, Day, Jacobson, Friesner, and Farid. "Novel Procedure for Modeling Ligand/Receptor Induced Fit Effects", J. Med. Chem. 49 (2006) 534. Online

We also have on-going collaborations with Jim McKerrow in which we apply these methods to targeting cysteine proteases in various parasites.   

ADME prediction

Many properties of small molecules in addition to their binding affinity to their target determine whether than can be used as drugs.  Recently we have become interested in predicting ADME properties.  Our published work thus far has focused on passive membrane permeability, in collaboration with Scott Lokey (UCSC).  We have developed an atomistic, physics-based model of passive membrane diffusion which appears to have a number of advantages relative to the statistical models in common use. 

Rezai, Yu, Millhauser, Jacobson, and Lokey. "Testing the conformational hypothesis of passive membrane permeability using synthetic cyclic peptide diastereomers", JACS, 128 (2006) 2510. Online

Rezai, Bock, Vong, Lokey, and Jacobson. "Conformational flexibility, internal hydrogen bonding, and passive membrane permeability: Successful in silico prediction of the relative permeabilities of cyclic peptides", JACS, 128 (2006) 14073. Online

 

Computational modeling of antibody-antigen recognition

Antibodies are another class of important therapeutic compounds.  Currently, computational approaches play very little role in the development and optimization of antibody therapeutics.  This project has several distinct but overlapping goals.  One major goal is to combine knowledge-based approaches (i.e., known "canonical" loop conformations) with state-of-the-art energy-based loop/side chain prediction to provide high accuracy models of antibody hypervariable loops.  The H3 loops hold particular interest for us, because they are much more variable in both sequence and structure, and have recently been shown to exist in multiple conformations upon binding different antigens.  We are also interested in the role of induced fit (primarily of the antibody) in antibody-antigen recognition.