Molecular Modeling - Seeing the Whole Picture with Modeling Software Packages

Date: August 31, 1998Table 1: Molecular Modeling Software, Table 2: Structure & Modeling Information and Table 3: Visualization & Image Resources The term "molecular modeling" usually conjures up two images: three-dimensional (3-D) depictions of small molecular structures (for example, benzene) and biological macromolecular structures (for example, proteins, DNA, and RNA). Some of the earliest and perhaps most renowned 3-D representations of a biological macromolecule were the wire-frame and paper models of deoxyribonucleic acid developed by James Watson and Francis Crick, who used this physical modeling technique to decipher the arrangement of deoxyribonucleotides in the DNA "double helix" (J.D. Watson, "The Double Helix," Antheneum, 1968). Within the next decade, computational tools had made their way into molecular structure analysis. Although crude by today's standards, these tools contributed to the structural analyses of cytochrome c and myoglobin. For an interesting historical perspective, read Cyrus Levinthal's "Molecular Model Building by Computer" (Scientific American, 214:42-52, 1996). Molecular modeling as a scientific discipline has advanced tremendously from these early years. Structure modeling is no longer confined to modeling gurus, and the use of multicolored Styrofoam balls, plastic straws, paper cutouts, or two-dimensional (2-D) images on monochrome monitors. Today, detailed molecular modeling can be achieved using basic desktop computer systems (Microsoft Windows and Mac) or high-end platforms (SGI, Cray) and a myriad of applications emphasizing various aspects of structure and structure determination. In addition, viewing and analysis of computationally derived models are not confined to 2-D (pseudo-3-D) computer monitors. Using a variety of peripheral tools and devices allows structures to be viewed in stereo or transformed into physical models. For example, specialized LCD glasses (such as Crystal Eyes [Stereographics Corp., www.stereographics.com]), an IR emitter interfaced with a stereo-compliant monitor, provides true 3-D viewing of displayed images. Physical models can be constructed using equipment such as the Helisys 1015 Laminated Object Manufacturing (LOM) device. (For additional information on LOM technology, visit www.sdsc.edu/tmf or read K.A. Svitil's "A Touch of Science," Discover, June 1998, pages 80-84). These tools enhance the entire analytical experience by allowing better depth perception of structures and the opportunity to utilize tactile senses to get a true feel for a given structure. In the instructional setting, protein and ligand physical models are excellent teaching tools in that they provide the means for students and researchers alike to experiment with various docking positions. Molecular Modeling Images provided by Molecular Simulations Inc. Because of the wide range of applications and methodologies, it's practically impossible to do them all justice in an overview of this length; therefore, this profile will focus on molecular modeling visualization applications for macromolecules (protein, DNA, and RNA). Specific methods will be discussed as they relate to functions that augment or are an inherent part of visualization and modeling. Our readers should also be aware that many programs typically thought of as "small molecule modeling applications" (for example, Chem3D [CambridgeSoft]) also are excellent macromolecule modelers, and vice versa. The Modeling Software and Suppliers Table (Table 1) lists a broad selection of modeling applications: from those without cost to the very expensive, and from basic visualization to multi-function, multi-method program suites. This is not meant to be an exhaustive list but a representative sample of what is available. Additional collections of software information are accessible on the World Wide Web; just point your browser to some of the modeling center sites listed in the Structure and Modeling Information Table (Table 2). Modeling in a very simplistic sense refers to the rendering in three-dimensional space (Cartesian) the atoms of a molecule, whether it be something as simple as methane or as complex as ribulose, bisphosphate carboxylase/oxygenase. How and what is rendered makes the field of molecular modeling very complex. Rendered images may reflect hydrophobicity and hydrophilicity of component residues, hydrogen bonding between distant residues on the same or different polymers, interior and surface electrostatic charges, space-volume constraints, and so on. Molecular modeling applications use falls into two broad categories: interactive visualization and computational analyses. The former involves the graphic representation of a structure and the subsequent subjective "by hand" manipulation of the structure. In this mode, the modeler's intuition plays a major role in structure determination. The latter category involves objective, computational analysis and is based upon known biophysical features of the molecule and established mathematical concepts that describe those features. These two approaches to modeling can be used alone or collectively to computationally derive a structure. Furthermore, these tools also can be used to reconstruct best-fit models from known structures when researchers make theoretical substitutions, insertions, or deletions in the composition of the macromolecule. Three of the most prominent uses of modern molecular modeling applications are structure analysis, homology modeling, and docking. Structure analysis centers on computational visualization of a molecule, provided its 3-D atomic coordinates have been elucidated, usually by X-ray crystallography or nuclear magnetic resonance (NMR). This information usually resides in major, world-accessible databases including the Brookhaven Protein Data Bank for protein structures, the Nucleic Acids DataBase at Rutgers University for DNA structures, and the Cambridge Crystallographic Data Centre (CCDC) for small molecule (nonprotein/DNA/RNA) structures. Refer to Table 2 for a list of structure information databases and search tools and their Internet addresses. Using structure analysis tools, investigators may dissect the intricate features of a molecule's structure or examine potential structural changes due to changes in the atomic or molecular composition of the molecule or macromolecule. For example, a molecular biologist may examine the effect of an amino acid residue change on structure and possibly function. An examination of the structure may reveal details not readily apparent from a 2-D analysis of a sequence. Three-dimensional structural analyses give the researcher the ability to examine the spatial, electrostatic, hydrophilic/hydrophobic, potential bonding, or the relationships of the substitute residue with neighboring residues on the same or separate chains. This structural analysis provides an avenue to examine numerous potential changes for "desired" effects, with the intent to select one or two for actual cloning experiments, saving the investigator time, resources, and expense. In addition, based on information gleaned from model analyses, the investigator may gain insight into how the cloned protein may behave during purification or characterization, and in kinetic analyses. Other uses of visualization include education and training. It is much easier today to depict a 3-D structure (especially if one has use of stereo functions with their application and hardware) than to attempt to describe this same structure. The old adage "a picture is worth a thousand words" has real meaning in this regard. Using visualization in conjunction with physical models, students of biochemistry and molecular biology can get a real grasp, a "look-and-feel" for a molecule, and examine how it may interact with a substrate or another protein (if docking partner models are also available). This was not possible a mere decade ago. Homology modeling has been very important these last few years, as researchers in academia and the pharmaceutical industry seek model structures for proteins whose crystal structures have not yet been solved. Homology modeling is also referred to as "comparative modeling" and "knowledge-based modeling." It is essentially the theoretical creation of a structure using structural elements borrowed from another protein within the same protein family (usually based on primary sequence and/or secondary structure features) whose crystal structure is known. The process involves alignment of the two sequences, usually performed by any of a number of bioinformatics tools (to be reviewed in the October 12, 1998, issue of LabConsumer). The result of this alignment is then fed into a homology modeling application, which uses the known crystal structure and the alignment to construct a "draft" (preliminary) structure for the "structureless" protein. This structure is then refined: loops are constructed (for "gapped" alignment regions), the residue side chain spatial placement is modified, and the entire "draft" homology structure is fine-tuned. A number of commercial and academic software packages perform homology modeling. Among the most widely used are INSIGHT II/HOMOLOGY and Modeler (MSI), Look/GeneMine (MAG), and SYBYL/COMPOSER (Tripos). Docking modeling is another area where major applications development and use have occurred over the last half-dozen years, as academics and pharmaceutical researchers seek to better understand and model novel protein-protein and protein-ligand interactions (that is, receptor and ligand binding). The latter has been extremely important in the pharmaceutical industry, as computational docking methods provide an avenue to examine and model receptor sites and assess potential ligands (drugs) and receptor-ligand associations. Using this computational approach, researchers can circumvent the long, tedious, and expensive process of bench testing each potential ligand. At the level of basic science, these techniques allow one to examine binding specificity and decipher the details of the atomic interactions involved in molecular recognition and catalysis. Some of the more widely used docking programs include AutoDock (Oxford Molecular Group), DOCK (Molecular Design Institute-UC San Francisco), FTDOCK (Biomolecular Modeling Laboratory), INSIGHT II (MSI), SYBYL/FLEXIDOCK (Tripos) and MidasPlus (Computer Graphics Laboratory, UC San Francisco). Important challenges in this area are optimizing the conformations of the ligand and receptor, and modeling the relevant non-bonded interactions between two species. The new docking programs GOLD (CCDC) and Flex X (Tripos) take the approach of applying data from X-ray crystal structures in the Cambridge Structural Database (see Table 2) which is a unique source of experimental information on non-bonded contacts. Other excellent, docking ("D" in the Modeling Functions column) programs are listed in Table 1. Although there is fair degree of latitude in subjective molecular modeling, at some point a researcher's "constructed model" will need to be refined--that is, computationally (mathematically) modified to comply with the known universal physical and chemical laws. In other instances, one may construct a structure model from scratch using primary sequence data and mathematical algorithms derived from physical laws and energetics concepts. Since these mathematical models have an inherent and important, albeit transparent, effect on modeling projects, a cursory discussion of their place in modeling is warranted. In essence, objective modeling revolves around three different approaches (each based on different underlying physical and chemical theories): molecular dynamics, molecular mechanics, and quantum mechanics. All of these are concerned with developing a unique solution to what is referred to as the "protein folding" problem--designing and testing algorithms and applications that will reliably predict 3-D structure from primary sequence data. The immediate benefit of these different lines of computational research has been the development, to varying degrees, of tools that allow us to refine subjectively created models or examine in intricate detail, and with high reliability, one specific feature of a model (such as, surface electrostatics). And one need not be a computational structure biologist to utilize these tools; many commercial applications for example, Cerius2, INSIGHT II (MSI), HyperChem (Hypercube), Look /GeneMine (MAG), SYBYL (Tripos)--provide transparent interfaces to these tools. Structural images created from X-ray and NMR coordinate files represent a snapshot in time for any given structure. In reality, the atoms and molecules are constantly moving as a result of thermal molecular motion and interactions with their environment. This interaction represents both passive and active processes (for example, interactions with nearby water molecules and substrate, respectively). In either case, these interactions translate into structural change for the macromolecule. The molecular dynamics approach to structure analysis seeks to understand and predict these structural changes based upon energy minimization. Dynamics analyses are based on an assessment (usually performed by molecular mechanics methods) of the free energy changes between two different structural states (a protein with and without bound ligand, for example). By mathematically extrapolating free energy changes, one can model a particular structure, which would have the appropriate calculated total free energy. These calculations can be reiterated for practically an infinite set of time points, thus allowing a researcher to model the temporal dynamics of macromolecule structure, for instance, as it performs some catalytic or binding function. Modeling programs that utilize molecular dynamics functions include HyperChem (Hypercube), INSIGHT II/Discover (MSI), AMBER, CAChe (OMG), SYBYL (Tripos), Alchemy 2000 (Tripos), Spartan (Wavefunction). Molecular mechanics emphasizes the potential energy of the molecule as a function of its component atoms, bonds and their angles, and charges--in general, the entire intramolecular environment. This approach attempts to calculate an energy potential for the entire molecule. Structure assignment is based on the assumption that the structure with the lowest energy potential represents a best fit for the molecule's structure as it exists in nature. Modeling programs that utilize molecular mechanical methods include HyperChem (Hypercube), SCULPT (ISI), SYBYL (Tripos), INSIGHT II (MSI), AMBER, CAChe (OMG). Quantum mechanics methods of "structure" determination are based upon the electronic makeup of a molecule. Electron distribution is defined by one of many quantum theories; the most widely known and used is the molecular orbital theory. This and other theories provide the basis for mathematically determining a number of physicochemical parameters (electric multipole moments of a molecule, electron density distriibution, electron affinities, nuclear atomic charge, electrostatic potentials, heats of formation, ionization energies, etc.) that may be utilized to construct a model structure. Quantum mechanisms offer advantages over the other methods in that it can be used to examine molecules at various electronic ("energy") states or during chemical bond formation and breakage. In essence, quantum mechanical methods are better at predicting chemical reality. The caveat to this statement is that these methods are generally restricted to analyses of small molecules. Nevertheless, they are very important in molecular modeling, especially when complemented by molecular mechanics methods. Modeling with quantum mechanical methods is best for detailed analyses of molecule surface electrostatics, and thus protein-protein interactions and active/binding site interpretations (structure-function relationships, mechanistics of catalysis and/or binding). Modeling programs that utilize quantum mechanical methods include Cerius2 INSIGHT II (MSI), HyperChem (Hypercube), and SYBYL (Tripos). The prospective user should keep in mind that many modeling applications make use of several mathematical methods; for example, Mac/PC/UNIX SPARTAN (Wavefunction), SYBYL (Tripos), and INSIGHT II (MSI) use a combination of molecular mechanics, quantum mechanical, and/or molecular dynamics methods. The novice modeler, researcher, student, or educator can get acquainted with structure visualization via a number of modeling tools--Web accessible and stand- alone programs. These applications are basic visualization tools and provide for no computational analysis of the structures displayed. Not to discourage a potential user from using a more sophisticated program, but not everyone has the need or the pocketbook for top-of-the-line applications. One can always graduate to a more sophisticated package and in most cases will be extremely pleased with the larger variety of routines and methods of analysis and visualization available. Some of the most widely used modeling applications in education are RasMol and Chem3D (Cambridge Soft). RasMol was developed by Roger Sayle and is available without charge. Using RasMol, as with many of the other basic programs (see Tables 1 and 3, Visualization and Image Resources Table), the modeler has the capacity to change what is viewed (part or all of a structure), how it is displayed (space-filled, stick-and-ball frame, ribbons, etc.), its color (as a function of B factors [atomic mobility]), etc. RasMol and another routine, Chime, may be configured as plug-in routines that work with web browsers to automatically display structures when the user accesses coordinate files. As Web-browser plug-ins, these are extremely useful tools in that a student or researcher can easily access and display structures on any computer, without regard to the type of computer system or operating knowledge of that system. Chem3D from Cambridge Soft is excellent modeling education tool. Its acclaimed organic/inorganic small molecule modeling functions are also applicable to macromolecules. Although not generally thought of as programs per se, Java-based Web applications (sites listed in Table 3) also are useful modeling tools in education. Many of them provide for the simultaneous viewing of structure and sequence, and like RasMol and Chime, they can be utilized from any computer that is connected to the Internet. Examples of other software used in the educational setting are Hyperchem (Hypercube), Alchemy 2000 (Tripos), CAChe (OMG), EduChamHS (Exceptional Teaching Concepts), MidasPlus (Computer Graphics Lab), XtalView (CCMS), WPDB (CCMS), ViewerProTM (MSI) and Spartan (Wavefront). The experienced modeler usually requires applications that provide some form of computational structure analysis along with visualization. They include applications that perform just one or two analytical functions or an entire suite of routines, complemented with excellent visualization tools. The former genre of programs is generally available from academic or nonprofit institutions for a minimal cost and includes programs such as DOCK, FLEX, PDBTool, XtalView, etc. These applications are particularly useful for those whose research interest is focused on one aspect of structure analysis. For example, a researcher examining of electrostatic surface charges of a molecule may utilize stand-alone versions of AMBER or MOPAC. Consult the modeling center sites listed in Table 2 for information on web links to these programs. For the modeler who requires detailed visualization coupled with a variety of analytical functions, integrated, multi-routine programs are available, such as Cerius2 (Tripos), INSIGHT II (MSI), Look/GeneMine (MAG), MidasPlus (Computer Graphics Lab), and SYBYL(Tripos). Programs of this genre are especially important to researchers performing homology modeling and docking, where various kinds of computational routines are utilized in the model-building process. Such a process incorporates all physicochemical properties into the computational equation to derive the best thermodynamically stable structure, a structure that should depict a functional molecule. The user should be aware that many of these programs require computer hardware more complex and expensive than a typical desktop system and that their learning curves can be rather steep. If these are not insurmountable hurdles, these programs will provide immeasurable opportunities for scientific discovery. Many modeling programs, but not all, allow modelers to annotate their models--that is, add text, navigation arrows, buttons, bullets, and so forth. Model annotation is an important feature if one wishes to use these models in a research publication (as a color print) or presentation (as a color transparency or slide). For programs without this function, the constructed image obviously will need to be annotated here. And even for programs with annotation capability, the format of the model electronic version may be incompatible with the printer or slide maker. These and many other post-modeling hurdles are easily overcome with any one of a variety of modeling accessory programs (most available at little or no charge). A brief selection is listed in Table 3. Another potential visualization method, currently in its infancy, is Virtual Reality Modeling Language (VRML). Although this method requires some degree of computer literary, it provides a novel look at molecular structure and is universally accessible. To use this technology, structure coordinate files are converted to VRML format and then viewed on any web browser with a VRML plug-in. VRML offers a unique advantage over other visualization methods: viewing is interactive. The modeler becomes part of the viewing experience, traversing the model, crawling down helices and barrels, making his or her way around loops or hopping over beta-sheets. For a comprehensive list of VRML tools and plug-ins for web browsers, point your web browser to the VRML Repository www.sdsc.edu/vrml/repository.html. Actually, we've come full circle, from Watson and Crick's wire-frame and paper models of DNA to the LOM-constructed physical models. What has changed in the interim is our fundamental understanding of molecules and macromolecules, the theories that explain molecular and macromolecular behavior, the algorithms that mathematically describe this behavior, the computational tools that can model this behavior, and of course, our access to and utilization of this body of knowledge and tools--all to the furtherance of our understanding of the biochemical world. Computation analytical tools, especially model building and visualization, will play larger and larger roles in biological scientific discovery, as technological advances in instrumentation generate, at an almost logarithmic rate, more data to be analyzed (c.f., advances in DNA sequencing and X-ray/NMR crystallography). For those who wish to know more about molecular modeling, a detailed theoretical description of molecular modeling is available in Andrew R. Leach's Molecular Modeling: Principles and Applications (Addison Wesley Longman Ltd., 1996); for an insightful look at computational protein modeling, read "Protein Modeling" (T.R. DeFay and F.E.Cohen, Encyclopedia of Molecular Biology and Molecular Medicine, vol. 5, page 158-69, 1996), or Protein Structure Prediction: A Practical Approach (M.J.E. Sternberg, ed., IRL Press, 1996). Table 1: Molecular Modeling Software, Table 2: Structure & Modeling Information and Table 3: Visualization & Image Resources Christopher Smith is an Associate Staff Scientist at San Diego Super Computer Center and can be reached at csmith@sdsc.com.

Molecular Modeling - Seeing the Whole Picture with Modeling Software Packages

Interested in reading more?

Become a Member of