The aim of our research is understanding the 3-dimensional
structure of proteins and how that structure is formed.
Our methods are the study and comparison of the known
structures, and especially the de novo design
of new model proteins and their synthesis and characterization.
We also develop new tools, where needed, for the
representation and modeling of macromolecules, such
as
- ribbon drawings for showing overall folds,
- kinemages and the associated Mage program for
discovering and communicating 3D ideas as open-ended
graphics on low-end (or high-end) computers,
- Sculpt for interactive modeling where real-time
energy minimization keeps the model physically realistic
while you tug on it, and
- the Reduce and Probe programs for improving the
actuality and accuracy of structure determinations.
Our study of the known protein structures has shown
features such as Greek key beta barrels, right-handed
crossover connections, beta bulges, helix N- and C-caps,
and Alacoils.
Protein design is a good way for students, trainees,
and all of us to ask very fundamental questions about
why certain sequences form the specific structures
they do, because starting all over again from scratch
can make you stumble over your wrong assumptions even
if you didn't know they were there.
The most interesting and general such result so far
is the discovery that protein design (and probably
folding and prediction as well) actually happens in
two stages and that the first stage is the easier
one. That easier first step, which we and other groups
have now done successfully a number of times, is achieving
the approximately correct tertiary structure. The
second, much harder step is achieving a unique, non-molten,
native-like structure with a well-ordered interior.
For both stages, negative design (actively avoiding
other alternatives) is at least as important as positive
design, and we believe that understanding the details
of packing will be crucial to the second stage.
Our lab has designed examples of many structure types
and has produced them by a variety of methods, to
strive for generality. Successful designs with approximate
but molten structures have included the 4-helix bundle
Felix (1FLX), the antiparallel coiled-coil Alacoilin,
the beta sandwiches Betabellin and Betadoublet (1BTD),
and the "TIM barrel" Babarellin. We developed
design rules for small SS proteins with variations
of the 2-SS, 56-residue SScorin design (1SSR).
Specificity is more important than stability for
native proteins to fold into unique, well-packed structures.
We developed a methodology to describe and study the
goodness-of-fit within proteins. All hydrogen atoms
are added and optimized, then a small .25A radius
probe is rolled over all atoms and all van der Waals
contacts, Hydrogen bonds, and physically impossible
atomic overlaps are scored. All-atom contact analysis
allows one to validate and filter macromolecular structural
coordinates to study just how real molecules pack.
Applied to RNA structures, including the ribosome,
resulted in a paper now appearing in PNAS that RNA
backbone is rotameric. Since all-atom contact analysis
also shows how model errors can be corrected, we are
combining this methodology with traditional crystal
structure fitting and refinement to produce more accurate
structures with both better fit to the data and better
geometry. The MolProbity server at http://kinemage.biochem.duke.edu
lets users evaluate their own, or their choice of
structures.
