We define computational biology very broadly to cover topics as diverse as gene annotation, protein folding, molecular orbital studies of biomolecules and their natural or synthetic ligands, ADMET modelling, modelling of protein folding, biomimetic design, computational drug, veterinary drug and agrochemical design and complex systems science.  The ability to model the interactions of biomolecules at the atomic level using sophisticated computational methods is increasingly underpinning advances in molecular biology, molecular pharmacology and molecular medicine.  Computational methods can increase the efficiency of design, discovery and development of commercially and medically important biomolecules such as drug, vaccines, protein therapeutics and gene therapies.  They will also be important in understanding important contemporary problems such as directing stem cell differentiation.  The products of such research will be extremely high value-added products, new knowledge, and new industries for Australia.

Australia has a long history in biomedical research and many current research groups in this broad area enjoy excellent international reputations.  Excellent computational biology groups have been formed across the country in the past two decades.  However, the connections and interactions between these groups are fragmented, and form on an ad hoc basis.  The object of the Network is to link and coordinate the efforts of these groups to improve their efficiency, capabilities, infrastructure sharing, and stimulate work in new areas where the groups overlap.  Many of the most innovative discoveries occur at these overlap areas.

The following material summarizes major focus areas and challenges in computational biology, with particular reference to strengths and capability in Australia (including the international CMSnet partners) which the Network aims to link: -
     

Proteins and macromolecules are the building blocks of living organisms. They provide the mechanisms for carrying genetic information, maintaining structural integrity and regulating the signaling network in living creatures, to name but a few of their roles. Such molecules, however, can be difficult to understand due to their size. They function at the interface of what can be described as purely microscopic molecular behaviour and macroscopic mechanical behaviour.

No single description of biological macromolecules can capture all the information we are interested in. For example, while a detailed ab-initio quantum calculation can yield important information about the interaction between two base pairs, it cannot describe the broader picture of how genetic information can be utilised and perpetuated through DNA. Thus, if we are to understand the function of biological macromolecules we must use a hierarchy of models, each of which is describing a different level of function.

The CMSnet will facilitate communication and exchange of new developments from different branches of computational chemistry, mathematics, physics and biology and bring this dynamic fusion of knowledge and computational techniques to focus on key landmark challenges in the area of systems biology. One such objective is to build layered, hierarchical descriptions of important biological molecules and to synthesise these into coherent pictures of the molecular structure, behaviour and biological function. This ambitious agenda for research in computational molecular science will in turn generate the type of quantitative data (based on rigorous molecular foundations) that is desperately needed for the development of reliable and predictive models of biological networks (e.g., genetic regulation, biochemical pathways).
　

Quantum chemical studies

While many of the properties of biomacromolecules can be studies by molecular dynamics calculations specific types of proteins such as enzymes, which catalyse the making and breaking of chemical bonds, require a different approach.  Chemical, excited state and optical processes require knowledge of the electronic properties of molecules.  Molecular orbital calculations are used to study the electronic properties of proteins and how these relate to their observed spectroscopic, electron transport, and chemical/catalytic properties.

The Biocomputational Chemistry  group at the University of Sydney use a small-molecule modelling approach with state-of-the-art ab initio quantum chemistry procedures to study problems of biological importance.  Their current interests are:-

• Reactions Mediated by Coenzyme B12: Determination of the mechanism of action of coenzyme B12 is one of the major remaining challenges in the B12 area. They are using an ab initio approach to tackle this problem.

• Peptide Radicals: Peptide radicals are implicated in both beneficial and deleterious biological processes. The latter include diseases such as atherosclerosis as well as ageing. We are investigating the stabilities and reactions of peptide radicals. The group are particularly interested in the consequences of oxidative damage to proteins resulting from the formation of peptide radicals.

• Carbapenem Antibiotics: Carbapenems are an important class of antibiotics because of their broad spectrum of activity and stability toward serine b-lactamases compared with the penicillins and cephalosporins. Although many carbapenems are natural products, for medicinal use they are produced by total synthesis. They are studying aspects of carbapenem biosynthesis because a better understanding may enable the development of more efficient production methods.

The Molecular Electronics computational research group at the University of Sydney employs a range of molecular orbital methods to study biomacromolecules. They use density-functional and other methods for the optimization of the structure and analysis of the function of large proteins. The size and complexity of most proteins requires the use of semi-empirical techniques for spectroscopic and transport properties of very large systems in excited states or non-equilibrium states. The group also use quantum non-adiabatic simulation methods for molecular and biological spectral simulation

Density-functional theory used to optimize the structure of green-plant photosystem PS-I (a 49000-atom protein)

Ion Channels

At the ANU a computational research group is studying ion channels.  The cell membrane, confining some ions and molecules and exchanging others with the environment, is the ultimate unit of physiology of life. The primary aim of their research is to understand the dynamics of ion permeation across membrane channels by applying rigorous physical principles and mathematical analyses. Their research effort should lead to a better understanding of many neurological, muscular and renal disorders, such as epilepsy, muscular dystrophy, cystic fibrosis and diabetes, as well as the action of new drugs for managing pains, various psychiatric indications and hypertension. Moreover, their work will lead to a better understand how a host of toxins and venoms interact directly with ion channels. Many of the potential pharmaceutical compounds being developed and tested are designed to interact with membrane ion channels. The techniques of modelling ion channels utilizing modern high-performance supercomputers can be readily extended for designing and screening pharmaceutical products.  The long-term goal of their research is to provide a comprehensive physical description of biological ion channels. Such a theoretical model, once successfully formulated, will be capable of predicting channel conductance from channel structure, and capable of revealing certain aspects of the atomic structure of protein macromolecules from observed conductance behaviour. It will link the structure and function of ion channels through the details of the inter-molecular potential operating between ions, water molecules and atoms that form the channel.

Biomaterials and surface interactions

The Physics Department at RMIT University in Melbourne has research interests in applying molecular simulation techniques to materials ranging from metallic, inorganic and hybrid systems to organic bio- and synthetic polymers and carbonaceous solids, with particular emphasis on surfaces and interfaces between technologically important materials. Members of this group pioneered a novel methodology for computer simulation of bio-molecular complexes and protein-surface interactions using Monte Carlo simulated annealing docking algorithm and molecular dynamics. 

Results of a fully solvated molecular dynamics simulation of the interaction of a petide helix with a protein surface.

Prediction of Structure and Folding

A collaborative program between the Centre for Computational Molecular Science (CCMS) and the Institute for Molecular Bioscience (IMB) at The University of Queensland is pushing towards macromolecular structure prediction with enhanced levels of accuracy. It is still very difficult to accurately predict with existing software and methods precisely how proteins, DNA or RNA will fold based only on their sequences. Currently new proteins are analysed by software developed to predict three-dimensional structure on the basis of sequence homology with other proteins for which three-dimensional structures have been experimentally determined. Although those analyses can often give a reasonable approximation to the real 3D structure, they are not of sufficient quality or accuracy to allow their use for such things as drug design. The CCMS/IMB team is exploring new approaches for structure prediction, which can be tested, for example on proteins studied within the IMB team (e.g. proteases and G protein-coupled receptors, the latter being membrane spanning proteins). 

 The UQ team are also developing computational models for protein folding dynamics. Understanding precisely how proteins (and other biomolecules) fold is one of the holy grails of biochemistry. Is protein folding template-driven or a concerted process from a molten globule? Potentially one can use in silico approaches to monitor and analyse folding pathways and mechanisms in real time. One can use various artificial tools in silico to probe such processes, such as the insertion of templates that maintain regions of desired conformation in an otherwise unfolded polypeptide. In a separate aspect of this problem, conformational dynamics are currently also difficult to predict. What makes a protein change structure?  For example amyloidogenic proteins associated with diseases like Alzheimer’s are known to undergo a conformational transition from helix to sheet. The development of computational methods for predicting the propensity for such changes is a key objective.

Collaborative work between chemistry and mathematics groups at Deakin University develops new methods for global optimisation. Current methods to find the very stable geometries or configurations of a protein or molecule can only give a probabilistic measure of the energy error in the configuration. The team at Deakin have developed a method, which will find the most stable configuration and give an absolute measure of energy error. In this way, it is possible to guarantee if the most stable configuration has been obtained or just a very stable configuration that may or may not be the most stable one.

The global pharmaceutical and biotechnology industries are some of the most substantial contributors to the world economy.  This contribution will grow as the population ages.  Paradoxically, the ability of these industries to develop new drug therapies appears to be steadying or even declining due to the increasing cost, complexity, competition and regulatory requirements.  In spite of new discovery paradigms such as combinatorial chemistry and high throughput screening being used, the number of new chemical entities (NCEs) registered each year has not increased.  Computational modelling and simulation methods provide the opportunity to increase efficiency of drug discovery and development, and essentially all major pharmaceutical, veterinary and agrochemical companies adopt computational approaches. 

Quantitative structure-activity relationships

The interactions of small molecules (drug and agrochemicals) with molecular targets, cells, organs and whole animals are usually complex, non-linear and unknown.  In many cases there is not sufficient detailed molecular information on targets and mechanisms of interaction to study these systems by molecular dynamics or molecular orbitals methods.  Pattern recognition, artificial intelligence and agent-based modelling methods are very useful in modelling these complex relationships.  Similar methods can also be used to build predictive models of other molecular properties such as water solubility, toxicity, or metabolism.  Many drug candidates fail late in the development process due to adverse pharmacokinetic or pharmacodynamics properties, metabolic degradation, side effects, or toxicity.  Companies are now using computational modelling methods of these types to allow such unsuitable candidate drugs to be eliminated earlier and at much lower cost (“fail early – fail cheap”).

CSIRO Molecular Science conducts fundamental research into the complex relationships between small molecule ligands and protein targets, or whole organisms.  This group has deconstructed the QSAR method and rebuilt it using optimum representational and mapping methods.  The group have devised novel molecular descriptors based on eigenvalues of molecular matrices derived from graph theory, and optimum structure-activity mapping methods using Bayesian regularized neural networks.  They have applied these techniques to modelling several important ADMET properties including blood-brain barrier partitioning, intestinal absorption, and acute toxicity.  Recent work carried out in conjunction with Medical faculty at Flinders University has resulted in novel methods for modelling phase II metabolism of xenobiotic molecules such as drugs.  Computational ADMET is also an important research area for the Medicinal Chemistry group at the faculty of Pharmacy at Monash University.

The IMB is also using computational informatics-based approaches to predict the bioavailability of drugs.  Oral bioavailability is currently estimated on the basis of polar surface areas and hydrogen bonding potential (e.g. J Med Chem 2002, 45, 2615-2623). However, data is available which could facilitate the development of a computational approach to ranking bioavailability. Similar methods are being employed to predict the blood-brain barrier permeability - a difficult property to predict.

Work at the University of Newcastle focusses on prediction of possible pharmacologically active compounds, developing computer-based 3D pharmacophores for use in database searches. Other work includes homology modelling of the kinase domain of receptor tyrosine kinases, analysis of X-ray structures and homology models to identify new drug target regions of low mutability, and computational studies of protein conformational changes. The figure to the left shows superimposition of the KIT crystal structure (red) onto the KIT_a models computed at Newcastle (green).

Complementary structure-activity relationships

These studies involve variation of the structure of both small molecule ligands and their large molecule receptor targets in order to better understand drug-receptor interactions. This work involves the interaction of medicinal and computational chemists, X-ray crystallographers, pharmacologists and molecular biologists associated with the Adrien Albert Laboratory of Medicinal Chemistry at The University of Sydney. Studies are directed at ionotropic receptors for the neurotransmitter GABA that are the site of action of a wide variety of drugs acting on the brain. A major aim of these studies is to design new agents that act specifically on subtypes of GABA receptors, including mutant receptors associated with epilepsy.

Structural biology, docking and scoring

Structural biology makes use of computers to take data from either NMR or X-ray crystallography and allow us to determine the three dimensional structure of both large (proteins) and small molecules.  The construction in Melbourne of a new synchrotron as a structural biology light source has provided additional focus for structure-based design of therapeutics.  Structure-based drug design starts with the information provide by the three dimensional structure of a protein and allows us to design molecules that will interfere with the normal action of the protein.  A structure also allows us to use computational techniques to probe the ligand-binding site to determine where chemical functional groups of various types are most likely to make favourable interactions with the site.  Structure-based drug design starts with the information provide by the three dimensional structure of a protein and allows us to design molecules that will interfere with the normal action of the protein.  The information in a protein structure can also be used to dock real or hypothetic molecules from databases into the site to determine whether are likely to be ligands.  Scoring functions allow the likely magnitude of binding affinity of docked ligands to be estimated computationally.  This allows us to locate likely lead structures more quickly than with random screening of chemical libraries.

Computational scientists and structural biologists at CSIRO Health Sciences & Nutrition/pHealth Flagship in Melbourne are studying Neurodegenerative Diseases. This is collaboration between CSIRO, Neurosciences Victoria, and the Dept. of Pathology, University of Melbourne. It is a multidisciplinary project involving molecular biology, protein chemistry, structural biology (cryoelectron microscopy, x-ray crystallography, and NMR), and quantum chemistry. The computational aspects of the project involve use of dedicated high-performance computing for the application of quantum and molecular mechanics to investigate reaction mechanisms and dynamics of proteins implicated in neurodegenerative diseases. Obviously, in an ageing society such as ours, study of the mechanisms of such diseases is becoming critically important.

Library design, diversity analysis and virtual screening

An exciting new paradigm for drug discovery, combinatorial chemistry and high throughput screening, was introduced in the 1990s.  This allowed some types of chemical synthesis to be performed in parallel and in an automated fashion, allowing the productivity of individual chemists to be greatly increased. It was possible to synthesis very large combinatorial chemical libraries using this method.  However, it soon became clear that the diversity of compounds in the libraries was most important for drug discovery, not the sizes of the libraries.  New computational techniques, such as cell based diversity metrics, were developed for measuring the diversity of libraries, comparing different libraries, identifying regions where new diversity was required, and for designing libraries with high diversity.  The computational analogue of combinatorial chemistry, virtual screening (a type of database mining) preceded the invention of combinatorial chemistry.  The development of new algorithms for rapid screening of extremely large virtual libraries (databases of chemical feasible, but hitherto unsynthesized molecules) is an active area of research.  Virtual screening encompasses both structure-based docking, and query-based, pharmacophore-based and QSAR-based approaches.

At CSIRO Molecular Science in Melbourne, a research team uses fingerprint-based methods to screen very large databases and libraries for novel antibiotics and veterinary drugs.  They have also developed new QSAR-based virtual screening and pattern recognition methods for bioactive design.  They are also actively working on using agent-based methods such as genetic algorithms and genetic programming to design focused chemical libraries with optimum target, ADMET and chemical novelty properties.

A collaboration between scientists in the Grid Computing and Distributed Systems (GRIDS) Lab. at the University of Melbourne and the Joint Protein Structure Laboratory, Ludwig Institute in Melbourne applies cluster and grid computing for virtual screening. Computational Grids are emerging as a new paradigm for sharing and aggregation of geographically distributed resources for solving large-scale compute and data intensive problems in science, engineering and commerce. However, application development, resource management and scheduling in these environments are complex undertakings. These researchers have developed a Virtual Laboratory environment by leveraging existing Grid technologies to enable molecular modelling for drug design on geographically distributed resources. It involves screening millions of compounds in the chemical database  (CDB) against a protein target to identify those with potential use for drug design. They used the Nimrod-G parameter specification language to transform an existing molecular docking application into a parameter sweep application for executing on distributed systems. They developed new tools for enabling access to ligand records/molecules in the CDB from remote resources. Computational researchers at the Ludwig Institute use such computational methods for the discovery and design of novel pharmaceuticals, particularly for the specific antagonism of protein-protein interaction. In particular they are interested in the development and application of novel scoring functions, and the use of data fusion methods to integrate dissimilar computational chemical design methods to yield improved scoring and virtual screening methods.

Molecular modelling-based bioactive discovery

The first 100 years or so of molecular modelling (the visualisation of three dimensional chemical compounds in our ‘real world’) utilised solid hand held metal or, more recently, plastic models. The rapid development of computer graphics over the last 20 years has enabled these nanoscale structures to be visualised and manipulated by the both the research and teaching communities with a reasonable degree of success. The fundamental problem lies in the fact that real molecular structures are three-dimensional yet the display screen is only two-dimensional and numerous computational ‘tricks’ are needed to try and represent three-dimensional topologies. Recent technological advances in both hardware and computer software environments have allowed molecular simulation to play an increasingly dominant role in the physical, chemical and biochemical sciences. A substantial number of research groups around Australia employ molecular modelling techniques to help understand the ways proteins interact with each other or with small ligands

Molecular modellers at Griffith University, Brisbane, have gone one step further. They simulate the real-time manipulation of atoms, molecules and macromolecular assemblies by employing tactile three-dimensional translational feedback mechanisms that informs the operator of, for example, the forces involved in the docking of a ligand into an enzyme or the affinity of an atom or molecule for a particular surface or binding interaction. One recent application of molecular modelling in a virtual reality (VR) setting accomplishes this haptic response by immersing the researchers in a computer-generated environment that synthesises molecular structure and function and has been successfully used to manipulate structural and mechanical properties of biological systems such as fibrin fibres in blood clots, adhesion forces in the ligand-receptor pair biotin and avidin, the viscoelasticity of cell membranes and even the mechanical properties of carbon nanotubes. The role of VR systems in the lucrative medicinal chemistry arena is proving to be invaluable. Currently the group are using software such as Amber 7 and NAMD located on the new Griffith University Sun cluster and our in-house Linux cluster to drive state-of-the-art virtual reality visualisation systems such as the program VMD via a commercially available Phantom haptic/Ghost software interface. They are utilising this resource in the study of:

• The dynamics and thermodynamics of assemblies of peptide nucleic acid probes and their RNA targets. These probes are the core components of a new range of novel commercially orientated biosensors.

• Specific mutations in the primary amino acid sequence of Green Fluorescent Protein (GFP) where we are carrying out theoretical calculations to predict variation in spectral characteristics and correlate these changes with mutant induced-tertiary structural variations.

• Investigate the de novo design of new materials for applications as chemotherapeutics for the treatment of pancreatic cancer and other associated cancers.

• To model the interaction of antisense dendritic macromolecules to their target gene sequences.

Computational immunology

Researchers at CSIRO Molecular Science, WEHI, and the Pharmacy faculty at Monash University in Melbourne have begun to apply some of the computational methods used for small molecule drug discovery to immunological problems.  The CSIRO and WEHI groups pioneered the used of neural networks to develop models of MHC class II binding peptide activity.  Previously these structure-activity models were derived using molecular docking techniques.  The QSAR and pattern recognition methods used by these researchers were quite successful in deriving explanatory models of peptide binding, and provided good prediction of binding affinity of peptide sequences not used in building the models.  The group at the Medicinal Chemistry department at the Victorian College of Pharmacy (Monash University) study recognition phenomena in interactions of immunological proteins.　

Computational systems biology approaches, particularly those aspects focused on understanding structure function relationship of proteins and ligands, have adopted essentially reductionist approaches.  By breaking systems down into their components, ultimately into molecular interacting elements, the premise is that a better understanding of the larger systems will result. In a Complex System, the interaction between the parts or sub-systems allows the emergence of global behaviour that would not be anticipated from the behaviour of components in isolation. This emergent behaviour depends upon the nature of the interactions as much as it does upon the character of the parts and changes when these interactions change. Such systems are inherently non-linear and so may exhibit hysteretic or irreversible transitions between alternative states. They are frequently characterized by fractal scaling laws and may exhibit self-organization. In recent years it has become clear that, due to the great complexity of biological systems, that many overt biological properties are emergent.  Recognizing this, and utilizing the tools of complex systems science, scientists are now beginning to study biological systems in a new way.  Australia has several research centres focusing on complex systems science approaches to modelling biological properties.

The CSIRO has a new initiative in CSS and supports several groups within the organization using complexity methods to study gene regulation systems, disease outbreak modelling, metabolic signalling, and complexity approaches to drug design. In addition, there are ARC-funded Centres for Complex Systems at the University of Melbourne, the ANU and UQ investigating biological systems amongst others from a complexity viewpoint. 　

Proteomics is the study of the way proteins work inside cells, and how they interact with each other. An important aspect of proteomic research is to study the differences in protein expression between diseased cells and healthy ones. This allows chemical markers to discovered which relate to the diseased state, providing a means of diagnosis. The work may ultimately lead to methods of gene therapy that will cure the problems at the level of the DNA in our genes. Glycomics is a new field involved with understand the role of carbohydrate-protein interactions at the cell surface in cell-cell communication. The proteome and glycome are the set of proteins and sugars respectively that an organism synthesizes. Sugars and proteins play a very important role in cellular function. Sugar are known to regulate hormones, organise embryonic development, direct the movement of cells and proteins throughout the body, and regulate the immune system.  

The Institute for Glycomics at Griffith University, Queensland, has research programs in the medicinal applications of carbohydrates. The Institute’s research profile covers areas such as cancer glycomics, microbial glycomics, including bacterial and viral glycomics, and the development of enabling glycotechnologies.  This group use Structural Biology and computational technique such as structure based drug design. The computational techniques used in the above fields includes molecular dynamics, visualization of the three dimensional structure of proteins and their substrates, docking of small molecules to proteins and the searching of large databases of commercially available compounds using in silico screening.

The anti-influenza drug Relenza (Zanamivir) shown in the active site of Influenza virus sialidase.

Protein dynamics
       

Biological macromolecules fold into defined functional shapes but their structures are not rigid.  Proteins undergo dynamic fluxional motions and it is important to simulate these to understand the relationships between protein molecular properties and their functions.  This is particularly true when studying the dynamic equilibria that occur when proteins interact with each other, or when a protein interacts with peptidic or small molecule ligands.  An understanding of protein folding pathways may enable prediction of protein sequence-function relationships or even prediction of tertiary structure. 

The MD Group at the University of Groningen (Alan Marks, international member of CMSnet) concentrates on dynamical simulation of biopolymers and lipid aggregates. Their aim is to understand and predict macroscopic behaviour of complex biomolecular systems on the basis of interactions between atoms. There are several levels of approach, all of which are pursued in the Groningen group:

• For processes where quantum-dynamical aspects are important, such as proton and electron transfer, a mixed quantum and classical dynamics is used.

• On the atomic level, classical molecular dynamics (MD) simulations are carried out on systems including up to tens of thousands of particles.

• On the molecular, but supra-atomic level, details of uninteresting atomic motions are averaged out or replaced by stochastic terms, thus concentrating on essential motions.

• On a ‘mesoscopic’ scale involving densities rather than individual molecules, classical density functional theory is used to describe dynamical processes that involve molecular aggregation and self-organization.

Fluorescent Proteins

Several research groups around Australia are studying the functionality, and dynamic and photophysical properties of Green Fluorescent Protein (GFP) using computational techniques.  The GFP and related protein fluorophores constitute an extraordinarily important platform for biotechnological applications. However, our present understanding of the photophysical properties this protein is almost entirely empirical, and computational methodologies for exploring the complex interactions of the chromophores and the enveloping proteins are at an early stage. Projects at the Centre for Computational Molecular Science (CCMS) at The University of Queensland utilize methods from quantum chemistry, quantum dynamics, and molecular dynamics to investigate the molecular basis of the photophysical properties of both the archetypal GFP and a range of exciting new fluorophores which emit in the very important red and near infrared parts of the spectrum.  The UQ group, recognizing that proton transfer is the key process that determines the remarkable properties of GFP, use ab initio quantum chemical methods to simulate proton transfer pathways between the GFP chromophore and the protein matrix at different levels of complexity.

Classic beta-can structure of a fluorescent protein. The chromophore species inside, largely isolated from water, is anchored by a helix which runs up the centre of the can.