Computational Structural Biology 
Axel T. Brunger, Ph.D. — Assistant Investigator 
Dr. Brunger is also Associate Professor of Molecular Biophysics and Biochemistry at Yale University. He 
was born in Leipzig, Germany. He received his diploma in physics at the University of Hamburg and his 
Ph.D. degree from the Technical University of Munich. He held a NATO postdoctoral fellowship and 
subsequently became a research associate with Martin Karplus in the Department of Chemistry at Harvard 
University before joining the faculty at Yale. His research has focused on molecular dynamics studies 
of protein structure and function and on methods in protein crystallography and nuclear magnetic 
resonance spectroscopy. 
OUR research lies at the interface between 
theory and experiment in the area of struc- 
tural biophysics. The research tools are simula- 
tion methods of computational chemistry adapted 
to the requirements of macromolecular systems. 
Macromolecular simulations are an important ad- 
dition to the arsenal of methods available to struc- 
tural biologists working with x-ray crystallo- 
graphic or nuclear magnetic resonance (NMR) 
spectroscopic data. In one class of projects, we 
are trying to understand the detailed microscopic 
interactions that govern stability and recognition 
in biological systems and to test the reliability of 
the theoretical methods as tools for this purpose. 
In another class of projects, we are directly com- 
bining macromolecular simulation with experi- 
mental data in order to make data analysis possi- 
ble or more efficient. 
Accuracy of Crystal and Solution 
NMR Structures 
As methods for determining macromolecular 
three-dimensional structure continue to become 
more powerful and are being applied to many 
biologically interesting systems, concern has 
been raised about the verification of final atomic 
models. A common problem arises when models 
are fitted against preliminary experimental data 
of mediocre quality. The recent revision of a num- 
ber of published structures, both x-ray and solu- 
tion NMR, illustrates the need to develop im- 
proved tools for checking the accuracy of the 
final atomic models. 
Structure determination of macromolecules by 
crystallography involves fitting atomic models to 
the observed diff'raction data. The traditional 
measure of the quality of this fit, and presumably 
the accuracy of the model, is the R value. Despite 
stereochemical restraints, it is possible to overfit 
or "misfit" the diffraction data: an incorrect 
model can be refined to fairly good R values, as 
several recent examples have shown. We recently 
proposed a reliable and unbiased indicator of the 
accuracy of such models. 
In analogy to testing statistical models by cross- 
validation, we defined a statistical quantity, Rjee. 
that measures the agreement between observed 
and computed structure factor amplitudes for a 
"test" set of reflections that is omitted in the mod- 
eling and refinement process. As examples show, 
there is a high correlation between Rj^e and the 
accuracy of the atomic model phases. This is use- 
ful, since experimental phase information is 
usually inaccurate, incomplete, or unavailable. 
The enhanced sensitivity of 11^,^^ with respect to 
model errors was illustrated for the crystal struc- 
ture of the plant ribulose-1 ,5-bisphosphate car- 
boxylase/oxygenase (RuBisCO) (David Eisen- 
berg. University of California, Los Angeles). A 
partially incorrect model for RuBisCO, which es- 
sentially had the small subunit traced backward, 
showed only a 4 percent conventional R value 
difference from the correct model. On the other 
hand, RX^g showed a 13 percent difference, sug- 
gesting that the incorrect model had been overfit. 
We concluded that Rj^g represents a reliable 
and unbiased parameter by which to evaluate the 
information content of a model produced by x- 
ray crystallography. It is not restricted to high- 
resolution diffraction data. The observation that 
Rjee cai^ distinguish between a random distribu- 
tion of scatterers and a distribution close to the 
protein suggests applications to ab initio phas- 
ing. Presently we shall apply this method to as- 
sess models of thermal motion and disorder, 
time-averaging, and bulk solvent in protein 
crystals. 
A similar approach might be useful for the 
three-dimensional structure determination by so- 
lution NMR. Using molecular dynamics refine- 
ment, we have just succeeded in implementing 
the free R approach. At present we are testing it 
for a number of model systems. The question of 
accuracy is an even more fundamental problem 
for solution NMR structures because of the ad- 
verse observable-to-parameter ratio. 
This work is also supported by the National 
Science Foundation. 
Predictions of Helix-Helix Association 
and Stability 
Prediction of the three-dimensional structure 
^ 7 
