We summarise below a number of key future directions for development of research in the field of Computational Molecular Science and its applications. Almost all of these directions will require advances in both fundamental methodology as well as application specific implementations. The dynamic fusion of expertise which CMSnet  represents and will foster to a much greater extent in the coming years will therefore be a key facilitator for advances in these fields which are crucial to the future of Australia's economy and society.

Materials and Nanotechnology
       

• Developments in theoretical methodology are required to facilitate many of the proposed technological applications. Below are some of the key areas where research is vitally needed:

• Efficient electronic structure calculations for large systems: Although it is already possible to perform linear-scaling calculations on molecules and solids, which represents the key to accessing thousands of atoms at the quantum mechanical level, there is considerable work still required for their optimal implementation, especially for parallel computers. There is also the need to explore means for improving the accuracy of the results through the introduction of localised post-Hartree-Fock approaches in condensed phases.

• Calculation of optical properties of solids: Because of the dominance of density functional methods in the solid state, it is not routinely possible to reliably determine optical properties ab initio, without adjustment. Work is required to explore the use of self-interaction corrected schemes to address the main error associated with the incorrect relative energies of atomic orbitals, particularly for virtual states. Further expansion of the use of Time-Dependent DFT is required for adsorption spectra, including the incorporation of electron-phonon coupling effects.

• Accurate quantum mechanical methods for condensed phases: While it is possible to perform very reliable ab initio calculations on small molecular systems, there are still many problems associated with methods that are feasible within periodic boundary conditions. The most widely utilised technique, density functional theory, has well documented failings, such as the failure to capture van der Waals forces for weakly overlapping densities. Research is required to find improved functionals, including hydrid ones, and new approaches to the solid state. This incorporates the further development of other state of the art advances, such as the GW, LDA+U, exact exchange, and optimized effective potential methodologies. All of the above can be benchmarked against quantum Monte Carlo calculations as improved sampling techniques are developed.

• Multiscale modelling: Many important processes, particular those in engineering materials, involve events on many different length scales that require different levels of accuracy in their treatment. For instance, crack propagation requires both the electronic detail of the breaking of bonds at the centre of the defect, as well as the continuum elastic response at long range. Integration of existing techniques from quantum mechanics through to computational fluid dynamics & finite elements is required to reliable treat such problems.

• Universal reactive force-fields: Atomistic simulation is considerably faster than quantum mechanical calculation and makes it feasible to sample a much greater extent of configuration space. However, the use of force-fields requires assumptions to be made about the minima to be sampled due to the inability to handle bonding changes in many approaches. Combining the advantages of bond-order potentials, for the short-range many-body component, with the important contribution of the long-range electrostatics and polarization will lead to a new generation of more reliable interatomic potentials that are applicable to most of the periodic table.

• Simulation of extended timescales: One of the greatest weaknesses of molecular dynamics as an approach to understanding the time evolution of a system is that it is only feasible to probe short amounts of real time, which is incompatible with the experimental processes being studied. Several methods for extending the timescale have been proposed, including hyperdynamics, basin hopping and kinetic Monte Carlo (KMC). However, there is still need for development to ensure that the acceleration of the time line does not alter the observation itself, as well as avoid incorrect preconceptions biasing the outcome, as can be the case for KMC.

Top

Biomolecular Modelling and Design
      

• MD simulations of larger systems. MD simulations are currently limited in size and timescale.  Better algorithms, faster computational resources, and the application of novel cluster and grid computing paradigms will enable more accurate MD simulations of biological systems.

• More accurate studies of protein function. Increased collaboration, resource sharing and the grid and cluster computing paradigms discussed above will enable large biological systems such as enzymes to be studied at a higher level of theory, yielding more accurate simulations and new insight into the structure-function relationships of biomacromolecules.

• Greater understanding of protein folding pathways. The ability to calculate the correctly folded tertiary structure of proteins from the sequence Is one of the holy grails of computational systems biology.  The application of new algorithms, forcefields, and novel complexity approaches will bring this dream closer to reality.  The increasing size of the protein 3D structure knowledge base, and the study of protein folding families, will enable other techniques such as threading and homology modelling to provide better computational models of protein structures not yet experimentally determined.

• Gene sequence/functions and protein structure/function relationships.The increasing protein structure and function knowledge base will allow novel computational methods to predict sequence-function and structure/function relationships with increasing accuracy.  This will accelerate the annotation of gene sequences and facilitate the discovery of new therapeutic targets for drug development.

• Methods for predicting ‘developability’ (ADMET) properties. Predicting the ‘drug-like’ properties of drug leads is a very high priority for pharmaceutical companies as it has the potential to save time and development costs.  Computational models to predict relevant ADMET properties are still fairly simple, and these methods and their resulting models will grow rapidly in their complexity and efficacy.

•  More accurate docking and scoring models for protein-ligand interactions. The interaction of proteins with small ligands is a very complex process to simulate computationally.  The flexibility of both target and receptor, and the nature of their interaction with each other make it very difficult to find the biologically-relevant binding mode in  a sea of false minima.  The complexity of the interactions of a ligand with a protein also makes the accuracy of scoring functions used to rank ligands correlate relatively poorly with measured binding affinities.  Improvements in docking algorithms, forcefields, scoring functions, and the impact of grid and cluster computing facilities will allow the usefulness of these models to improve substantially.  This should allow give us and enhanced ability to simulate the much more difficult protein-protein interaction problems.

•  Computational phamacogenomics. The development of computational systems biology methods will enable the identification of potent specific targets for regulatory networks. Current research indicates it is the overall structure of such networks, rather than individual rates for components, that is important. This changing viewpoint will impact on the nature and targets of therapeutics. The impact of pharmacogenomics will require variants of therapeutics for “tailoring” to different patient types. Computer aided design methods will be required to deal with these demands.

•  Improved virtual screening methods. The development of virtual screening methods to rival HTS methods is required. Such methods will require advances in homology modelling and protein structure prediction (in both apo and bound forms) binding energy prediction and toxicity modelling methods.

• Haptic methods. The medicinal chemist of the very near future will not only be able to ‘experience’ three-dimensional environments such as enzyme active sites and receptor substrate binding domains, they will be able to interactivity dock substrates into these sites and sample a number of ‘what if’ binding situations quantitatively, visually, and tangibly. The power of these systems is that the operator can temporarily assign zero weight to forces and ‘physically’ move a model through of round energy barriers that are impossible in ‘conventional’ molecular modelling techniques.

• Development of a priori MD simulation techniques for higher accuracy and increased applicability to, e.g., metalloproteins and optically or electronically excited states.

Top

• It will be important to continue to develop more efficient methods for the construction of ab initio potential energy surfaces for chemical reactions. This development will be paralleled by studies of important reactions which are the subject of intense investigation in Australia and overseas.

• However, two new areas will dominate the research. Firstly, many chemical reactions, particularly in photochemistry, involve molecular motion in more than one electronic state. To study these reactions from first principles, the ANU group is developing a method to construct the ab initio diabatic potential energy matrix that governs the molecular motion. These reactions will be studied with state-of-the-art quantum dynamics methods in collaboration with the National University of Singapore. Secondly, the study of large molecular systems and reactions in solution will require a many-body-expansion approach to the potential energy surface. Methods and computer codes to implement this approach will be the subject of research in the immediate future.

• A major area of future development in molecular quantum scattering will undoubtedly be the pursuit of a new generation of iterative methods which go beyond the traditional wavepacket paradigm of "one propagation = one column of the S-matrix. The Lanczos based technique introduced recently by the CCMS at UQ is presently the best candidate in the international arena for upscaling to difficult multidimensional problems.

Top