University of Alabama at Birmingham
Center for Macromolecular Crystallography
Box 79 THT, UABCMC, B'ham, Al 35294 (USA)
phone: 205-934-5329 fax: 205-934-0480
carson@luna.cmc.uab.edu or carson@uabcmc.bitnet
The IRIS Inventor is an object-oriented 3-D graphics toolkit developed by Silicon Graphics. This C++ programing environment provides a wide variety of geometric primitives and methods for rendering them, including such advanced features as true transparency and texture mapping. These rendering features have been used to illustrate the following in the first two weeks we have had the software, by extending the Ribbons 2.0 program: molecular replacement solution of a protein structure, location of regions of potential coordinate error in the structure, and the results of the successfull structure-based drug design of PNP inhibitors initiated in this laboratory. The more advanced 3-D manipulation and event handling actions provided by the toolkit will be explored in the months leading up to the meeting. Potential applications include the re-fitting of a partially refined structure against the X-ray data, and the docking of inhibitors into an enzyme active site. The latest results will be presented.
(The abstract was published as part of the annual meeting report: J.Mol.Graphics 12:67.
It's not actually part of the poster.
Below is a web-based recreation of the poster given at the MGS'93 international
meeting in Interlaken. The actual poster decorates the graphics lab at the
CMC.)
Malic enzyme crystals grown in microgravity show dramatically improved) diffraction quality. (data courtesy of Howard Einspahr and Larry Delucas).
HKL data rendered as cones aligned along their reciprocal lattice vectors, color-coded and scaled according to F(observed).
Thresholded slab of data in c* with transparent spheres setting low and high resolution limits. Reciprocal lattice of major zones shown.) show
A landscape portrait of the diffraction pattern from X-rays illuminating space-grown insulin crystals.
Insulin diffraction intensities collected on a Siemens area detector (data courtesy of Craig Smith and Marianna Long.) A rotation photograph created by averaging 10 consecutive frames of the 512 x 512 pixel sampling of the raw data.
Image created with IRIS Explorer, SGI's data-flow visualization environment. Peak height proportional to sqrt Intensity. Colormap based on standard deviations above background, adjusted to make the "water ring" blue.
Geometric/graphics models created by the IRIS Explorer may be imported into the IRIS Inventor.
The artwork "Diffraction Space: The Final Frontier" was composed with the "land" as a diffraction pattern rendered and viewed from an interesting angle. Insulin as a space-filling model is the "moon".
The image illustrates some of the awesome power of computer graphics: --- None of my colleagues could guess what the image really was!!! The same data viewed from a more conventional perspective was immediately recognized from the circular area of the detector and the beam stop.
Amazing Fact: Space-grown insulin crystals average three times the intensity of earth-grown crystals of the same size (Long et al, submitted).
PNP (Purine Nucleoside Phosphorylase) degrades anti-cancer drugs resembling purine bases and is thus a target for structure-based drug design. PNP has been extensively studied at the CMC and BioCryst (data from John Montgomery, Charlie Bugg, and many others.)
Methods: Iterative repetition of: look at enzyme/inhibitor complex; synthesize new inhibitor; test chemically/crystallographically. Results: nanomolar level inhibitors. 1000x improvement (PNAS(91) 88:11540.)
Graphic employs green and yellow ribbons for the subunits comprising the active site (~90%) in green). Key active site residues drawn as ball and stick models. Transparent surfaces highlight substrate guanine (blue), ribose (white), and phosphate (magenta). Textured aromatic ring plane has a dart object imported just for fun.
Renderings of crystal packing of R32 cell treated as hexagonal.) show (PNP trimer packs as three hexamers per cell, a=b=114, c=315 A) show (Huge solvent channels allow easy access to the active site.) show (Enzyme is fully functional in the crystalline state.
Cell packing diagram gives perspective view of many unit cells (yellow lines)) show (and the symmetry related PNP monomers (blue textured ellipsoids).
Solvent channel close-up with symmetry axes (yellow lines) and PNP) show (monomers (blue "shrink-wrapped" surfaces).) show (Guanines (reddish, textured ellipsoids) are near the active site entrances.) show (A semi-transparent plane representing solvent slices through the active site.
Model structure built by homology with serine proteases. Self rotation function, polar form, kappa=180. Cones as potential pseudo-2-folds with a red line marking dominant peak and unit cell vectors superimposed.
Cross rotation function, Eulerian angle space as 3-D grid. Axes are placed on two predominant peaks rotated by given theta-1, theta-2, theta-3 (related by a 2-fold.)
Translation function, R-factor minimum search / packing diagram. Blue molecule held at origin (arbitrary in P1) nearest to peak. 2-fold related cyan molecule centered at peak (semi-transparent cubes.)
Factor D is the rate-limiting enzyme in the alternate complement system. The structure has been solved by MIR methods and refined to R=0.18 at 1.8 A resolution (Narayana et al, submitted to JMB.)
Initial model built by homology with "Atom" (my FRODO variant), molecular replacement unambiguous, X-PLOR refinement gave R=.25 at 2 A Procedure later repeated with Quanta's commercial homology building package. (Note: these models are "untouched by human hands" during refinement.)
Final refined structure allows comparison to models and assessment of errors. Ribbons show model deviations color-coded from pale blue to red in 0.5 A steps. Left side are homology models, my(M) and Quanta's(Q). \ Right side after Xplor. Models M and Q deviate equally from each other and from final structure. X-PLOR refinement against experimental data places 90% within estimated error.
The errors are localized in the polypeptide substrate binding loops --- thus the modeling only got the most critical residues wrong.
Various metrics have been proposed for error detection in X-ray structures. The refined Factor D model satisfies these metrics, and is taken as correct. The deviations between the MIR model and partially refined homology models are taken as the error function and correlated with the various metrics.
View PostScript plot of Correlations.
The metrics used are real-space R-factor (omitmaps and 2Fo-Fc), B-factors, sliding R-window, geometric strain energy, phi/psi "energy", rms shifts in refinement, deviations from database, exposed surface, chi angle "energy", 3D folding profile, and omega. (analysis from "O", "XPLOR", Ribbons.)
Inspection and computational considerations imply the 5 best are: B-factor, real-space R, dihedral "energy", rms shift, and geometric energy. Statistical analysis suggests a linear model of these 5 variables. Grossly incorrect residues may then be flagged with 90% accuracy.
View PostScript plot of Error Analysis
Inventor allows image files to be texture mapped easily to surfaces. Textured views of the long helix of gamma interferon are shown. Red/blue textures on the back side mark amide/carbonyl direction, flowers and wallpaper alternate for the residues (top). 3-D letters are available also, providing another way to show sequence.
Inventor provides a variety of direct 3-D manipulator objects. The prototype Ribbons++ viewer is shown below with a histidine molecule. A spaceball manipulator is shown attached to the bond to be rotated. One should be able to extend their objects to those suitable to molecules.
DNurbs (dee-in-urbs): a small number of bicubic patches wrapping each base pair in complementary DNA (submitted to J.Mol.Graphics).
Images employ 8 B-spline patches, colored by base type with sugar/phosphate magenta, major groove white, minor groove silver. Cutaway views with stick bonds and transparent major groove (left). Textured patches with sequence letters and chain direction arrows (right).
Inventor is a powerful system for composing graphics. Using advanced graphics techniques such as true transparency and texture mapping is made easy.
C++ programming is harder++ than C. This is my first attempt at C++, after 10 years experience with C. Don't quite have it yet (after 10 weeks), but I like it so far.
Inventor is a large and complex object-oriented environment. It is often difficult to discern relationships between classes, yet the system is well-documented and a good way to learn C++.
The goal of a serious interactive program was not attained. Ribbons++ does not have nearly as good an interface as Ribbons. It has proven very difficult to extend the interactive classes. Don't expect much help on this from the SGI hotline. This is summarized in the cartoon below:
Future work: Create an interactive system for re-fitting proteins. Quote from an anonymous group member, after hours on the graphics: "Sometimes I feel like a crane operator."
But this must be made fun! Like Super Mario Brothers.
prototype Ribbons++ viewer