Broadly defined, computational biology is the use of computers to solve biologic problems. Richard W. Pastor, Laboratory of Biophysics, U.S. Food and Drug Administration (FDA) Center for Biologics Evaluation and Research, Rockville, MD, sees computational biology as encompassing several fields, including bioinformatics, systems biology, and his own area of research, molecular dynamics simulation. “How do you analyze all the data in the huge databases coming out?” he asks. “The tools needed include high-powered statistics, e.g., Bayesian statistics and high-powered computer searches.” But computational biology is not just about analysis for its own sake. “What do people want to know about? Diseases people have. Going back to systematics and reanalyzing evolutionary trees, seeing where branches split. We cannot only go back with new tools and data, but go forward to medical applications. There are questions from 100 years ago we can now solve.”
Molecular MachinesAndrew Neuwald, Associate Professor, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, performs computational analyses of proteins that are well conserved across major groups of living organisms. Certain residues have been conserved for over a billion years, suggesting they have critical functions. He has developed procedures for classifying and estimating selective constraints on specific residues and analyzing corresponding molecular interactions. Although proteins in large classes of well-conserved families (e.g., the AAA+ class of ATPases, signaling GTPases, and protein kinases) are most amenable to this analysis, he also uses it for smaller classes (e.g., DNA sliding clamps).
“We've had an industrial revolution in biology,” notes Neuwald, “with experimental factories creating massive amounts of experimental data which is not hypothesis-driven, including the genome project, microarrays, and two-hybrid experiments. A fundamental question is how to maintain scientific rigor and apply the scientific method rather than ad hoc approaches.”
Neuwald asks, “How can you apply a scientific method approach to non-hypothesis-driven science, like looking at mountains of sequence data? I think to ensure scientific rigor, you need to employ Bayesian statistics.” Rather than assume a hypothesis is true and trying to disprove it, this approach considers every hypothesis at once and doesn't assume what the true model is. In fields such as systems biology and signal transduction, the single hypothesis approach breaks down as an unfeasible number of possible models could be set up and results consistent with the hypothesis could be false.
For Neuwald, proteins are the simplest biologic system one can study. “All the residues impart some function, but the system is something more than the sum of its parts. The scientific reductionist approach is to isolate components and see how they work, but this doesn't work for proteins. Bayesian statistics can infer and take into account properties of the system.”
“Bayesian statistics gives a blurry probabilistic view of a system. Physicists working with quantum mechanics are used to thinking this way,” Neuwald observes. “Biologists have to start thinking this way. The role of statisticians is underappreciated.”
Modeling MembranesPastor's work is based on the observation that Newton's equations of motion, used to describe the orbits of planets around the sun, can also describe all motions in fluids other than chemical reactions or motions in superfluids. To create molecular dynamics simulations in biologic systems, all atoms in a membrane are treated with Newton's equations of motion solved for a certain length of time. Atypical simulation involving 20,000 to 50,000 atoms requires solving about 100,000 differential equations. “These simulations can take 3 months on pretty big clusters of computers. We are looking at 10 to 15 nanoseconds of the process, and the membrane probably takes a microsecond to fuse, and maybe a second to fold, so we are looking at early stages,” Pastor says.
Image 1.
Some of his group's work involves looking at the interaction of influenza virus hemagglutinin fusion protein interaction with the cell membrane. Because of its polar residues, the influenza virus hemagglutinin fusion peptide can only partly bury itself into the top leaflet of the lipid bi-layer. To avoid an unfavorable vacuum region, the lipids’ adjacent lipids curl under the peptide, while those on the opposing leaflet extend into the upper leaflet. They speculate that these events trigger positive membrane curvature and eventual cell fusion. A long-range plan is to simulate bigger membranes and look at how the virus fuses, eventually testing drugs to prevent fusion. Other applications are simulating protein folding.
