Molecular Modeling in the Genomics Era

Molecular Modeling, Visualization, and Structure Prediction Software (Additional material not included in the print edition) Courtesy of Theoretical Biophysics GroupBeckman InstituteUniversity of Illinois at Urbana-ChampaignThe photosynthetic reaction center of the purple bacterium, Rhodopseudomonas viridis (rendered using VMD) The aim of some life science researchers is to understand human physiology and the various diseases and mutations that can cause the body to go haywire. Accomplishing this goal requires a thorough understanding of the workhorse of all biological systems, the protein. Currently, a tremendous amount of effort is spent on proteomics research as scientists map protein-protein interactions (proteomics tools and technologies will be profiled in LabConsumer on April 2, 2001). Yet there is also a push to map the structural features of proteins, and the structural basis of protein-protein, protein-ligand, and protein-drug interactions. This effort requires the development and enhancement of software capable of carrying out molecular modeling, structure visualization, structure prediction, and protein folding and energetics calculations. Research, in silico In the early years of molecular modeling and protein structure analysis, software tools were created to aid both the subjective refinement of protein structure (coordinate maps) and the simple visualization of macromolecular structures. These early software packages were supplanted by complex applications that gave researchers the tools to explore solved protein structures in several dimensions. These tools also provided novel opportunities to examine the relationships between a protein and its substrates, inhibitors, and other proteins. Today, structure examination tools, in addition to providing the most advanced visualization features, must also offer analytical and predictive functions. Advanced analytical functions provide a mechanism to examine in detail the specific motifs, domains, and folds of proteins and to make comparisons against other proteins. Predictive methods allow researchers to gain an understanding of the possible structure and function of proteins whose existence is only hypothesized based on genomic sequence data, and which are not even known to exist otherwise. Most prediction methods can be grouped into three general types: homology modeling, two- and three-dimensional structure prediction, and ab initio (structure folding and energy minimization methods). A variety of molecular modeling tools and basic modeling concepts were reviewed in a prior LabConsumer molecular modeling technology profile.1 Many of these applications are still available, though more recent versions have been released with new and enhanced features. While the modeling basics have not changed in the past two years, massive sequencing efforts have resulted in the publication of the complete genomes of several eukaryotic organisms: Caenorhabditis elegans, Arabidopsis thaliana, Drosophila melanogaster, and Homo sapiens. The recently published draft sequences of the complete human genome, by the International Human Genome Sequencing Consortium and Celera, contains an estimated 26,000 - 38,000 genes encoding a comparable number of proteins.2,3 To date, only a small fraction of known proteins, and even fewer of the number hypothesized based on genome sequencing data, have had their structures solved by classical crystallographic methods or by nuclear magnetic resonance (NMR) spectroscopy. Although a few technological advances have occurred to enhance crystallographic techniques, including phase discrimination for multi-wavelength synchrotron X-ray diffraction data, better data collection, and improved computational interpretation of coordinate maps, these methods cannot keep pace with the demand to know the structures of all proteins, especially those postulated to be involved in medical maladies. The new genomics era has been accompanied by a radical shift in the kinds of questions being asked in basic biology and medicine. Researchers, armed with life's basic blueprints, can now investigate many of the "what if" questions that were previously unanswerable. This shift has been particularly powerful in the field of medicine and in the intensive search for new pharmacophores to aid in the treatment of human diseases and ailments. The foundation of new drug development efforts is now based on a thorough understanding of the relationship between the drug and its target protein, whether it is an enzyme, receptor, structural, or transport protein. Detailed information regarding the physical, chemical, biochemical, genetic, and physiological characteristics of that relationship must be obtained. Hence, the enormous push to enhance the pace and resolution of classical crystallographic and NMR methods for structure determination, and to improve methods for predicting protein structure. The latter goal represents a paradigm shift in structural biology that has occurred during the genomics era. While crystallographic and NMR techniques still receive a major share of the experimental and intellectual effort, structure prediction studies have been afforded newfound stature, and have received new impetus from the development and refinement of new models and methods. Increased functionality CambridgeSoft's Chem3D software Molecular modeling serves several functions in research today: it provides a mechanism to visualize structure in three dimensions; allows researchers to analyze elements of structure and make protein-to- protein comparisons; predicts the structures of proteins; and facilitates computer-aided drug design (CADD). The visualization function enables researchers to change their perspective of a protein structure, either by rotating the virtual three-dimensional model in the X, Y, and Z planes, by changing the display characteristics (e.g., from surface features to the Ca carbon chain), or by changing how the model is displayed (e.g., ball-and-stick, space filling, or ribbon formats). Another key feature of the visualization function is image output. As more researchers investigate the protein underpinnings of biological phenomena, there is a growing need for exact, detailed, and colorful images that can be incorporated into research publications and reports. What differentiates applications is the complexity of features and the extent of interactive visualization. Basic visualization tools and full-featured visualization packages are both listed in the accompanying, Web-accessible table. The World Index of Molecular Visualization Resources (www.molvisindex.org) contains an interactive listing of visualization products where users can add new tools or modify the descriptions of tools that are already indexed. These tools provide novel opportunities for researchers to experiment with protein structure and function and to become a part of the visualization experience, traversing the virtual model. Structure analysis is a major tool for structural biologists. With the development of user-friendly interfaces and documentation, a growing number of biological scientists can capitalize on the wealth of structure analytical functions provided by molecular modeling tools. The predominant analytical functions include comparative tasks, homology modeling, and docking. Basic comparative functions include such routines as calculation and mapping of surface charges on a protein, identity of common structural features and motifs, and comparison with other proteins. Homology modeling facilitates the creation of the hypothetical structure for an unknown protein using as a template another protein that shares many of the sequence features (>30 percent identity) and whose structure is known.4 Docking involves tools to model both protein-protein and protein-ligand interactions. Molecular modeling applications that provide an excellent selection of these analytical functions are also listed in the accompanying table. Courtesy of GaussianA ball-and-stick model of angiotensin Primary predictive functions include two- and three-dimensional structure prediction, threading, and ab initio methods. Secondary structure prediction is rather straightforward, albeit imperfect, and has not changed dramatically since development of the Ramachandran plot. How secondary structure elements such as coils, beta sheets, and turns are folded into a three dimensional "native" protein structure presents another set of complex problems. For proteins whose sequence identity is less than 30 percent, structure prediction is confined to ab initio folding and threading techniques. Ab initio methods build a protein structure by calculating the energetics of protein folding and of the protein in its final, native folded state, using the primary amino acid sequence as the initial building blocks.5 These methods use one or a combination of molecular dynamics, molecular mechanics, and quantum mechanics techniques to describe atomic and molecular states and interactions. The advantage of this atomic building block approach to the molecular landscape of a protein fold is that it calculates a structure that conforms to universal chemical and physical laws. Threading is essentially a small-scale form of homology modeling based on the realization that there are a limited number of structural folds that define the structure of a protein in its native, most energetically favorable state.6,7 It is predicted that only a few thousand distinct domain folds exist for all proteins and that all of these folds can be revealed by examining approximately 10,000 unique proteins. The caveat, of course, is that the 10,000 proteins must be carefully chosen to ensure representation of proteins where novel folds are mostly likely to be found. The individual folds serve as precise descriptors in protein structure prediction software, and aid in the development of new structure prediction algorithms, while the resulting collection of protein folds is a essentially an encyclopedia that describes each of the possible structural motifs a protein may adopt. These fold libraries are used by threading algorithms to create the theoretical structure of a protein.8 The process is akin to building a puzzle, using the folds as pieces and a sequence-fold map as the box-top picture. Courtesy Numerical Algorithms GroupDisplay of a buckminsterfullerene molecule with IRIS Explorer 5.0 Because structure prediction techniques analyze millions of atoms and calculate atomic movements, they are computationally intensive, requiring fast computers. In the past, this meant purchasing mid- to high-end computer platforms, but the introduction of high-speed CPU chips (exceeding 1,000 MHz) and dual-processor desktop machines provides new modeling opportunities to scientists with modest research budgets. The final molecular modeling application, computer-aided drug design (CADD), involves the design of potential drugs based on structure, physical, and chemical constraints and parameters between the drug ligand and its target, a protein. This "rational drug design" methodology has been a very powerful tool in pharmaceutical research, where drug development has traditionally been a hit-or-miss process of blindly testing compounds that may or may not interact with a protein. Using CADD methods, researchers can now model drugs to a specific set of criteria. When combined with combinatorial chemistry methods, scientists can dramatically increase their chances for a hit. Recent advancements in CADD include the introduction of evolutionary techniques, such as genetic algorithms and fitness functions, to CADD modeling.9 Genetic algorithms ind-uce spontaneous atomic changes to create new molecular species, while fitness functions ascertain the chemical survivability of the new species amongst the pool of biochemicals. Courtesy of TriposA ribbon diagram Composer model (based on GeneFold's sequence alignment) of 1eaf superimposed on its experimentally derived structure Prediction method development does not occur in a void, without some measure of applicability to the real world. The Critical Assessment of Techniques for Protein Structure (CASP) conference provides a forum where researchers can use their particular methods to predict the structure of a protein whose X-ray resolved structure is soon to be published. This provides a true blind assessment of the reliability of prediction methods, which have in the past been tested largely based on their ability to accurately predict the structure for a protein whose structure was already known. For more information on molecular modeling, CASP, and prediction methods, see references 10-14. S Christopher M. Smith (csmith@sdsc.edu) is a freelance writer in San Diego. References 1. C.M. Smith, "Molecular modeling," The Scientist, 12[17]:17, August 31, 1998. 2. J.C. Venter et al., "The sequence of the human genome," Science, 291:1304-51, Feb. 16, 2001. 3. International Human Genome Sequence Consortium, "Initial sequencing and analysis of the human genome," Nature, 409:860-921, Feb. 15, 2001. 4. A. Sali et al., "Evaluation of comparative protein modeling by MODELLER," Proteins, 23:318-26, 1995. 5. A. Ortiz, "Ab initio folding of proteins using restraints derived from evolutionary information," Proteins 37(S3):177-85, 1999. 6. S.E. Brenner et al., "Population statistics of protein structures: lessons from structural classifications," Current Opinions in Structural Biology, 7:369-76, 1997. 7. C.A. Orengo et al., "Protein superfamilies and domain superfolds," Nature, 372:631-4, 1994. 8. R.T. Miller et al., "Protein fold recognition by sequence threading tools and assessment techniques," FASEB Journal, 10:171-8, 1996. 9. A. Globus, "Automatic molecular design using evolutionary techniques," Nanotechnology, 10:290-9, 1999. 10. N. Guex et al., "Protein modeling for all," Trends in Biochemical Sciences, 24:364-7, 1999. 11. A. Leach, Molecular Modeling: Principles and Applications, Addison Wesley Longman Ltd.: Glenview, Ill., 1996. 12. CASP, predictioncenter.llnl.gov 13. D.M. Webster (ed.), Protein Structure Prediction: Methods and Protocols, Humana Press, 2000. 14. C.M. Smith, "Bioinformatics, genomics, and proteomics," The Scientist, 14[23]:26, Nov. 27, 2000.

Molecular Modeling in the Genomics Era

Interested in reading more?

Become a Member of