Modelling and Software
Modeling in the physical and life sciences encompasses a wide range of approaches that address equilibrium, and to some extent, nonequilibrium states, which allows one to address behaviors ranging from multi-phase fluid flows, blood flow, Brownian motions describing macromolecular diffusion, short-time rare transitional events in molecular systems. Today, those involved in modeling are more frequently confronted by increasingly complex systems requiring more general models. This, in turn, translates to more complex models, in which one can expect to find approaches encompassing multiple time and size scales, as well as models that incorporate multiple physical models, as well as incorporating experimental observables. In turn, this will require advances in algorithmic developments permitting the implementation of these new models on modern computer architectures.
Stochastic non-linear systems
The growing complexity of flows at the point of transition to turbulence has often been interpreted as the onset of a deterministic chaos. However, many flows do not present experimentally observable bifurcations. This drew attention to the possibility of flows to be excited by stochastic perturbations. In its simplest form, the mathematical model describing this situation is that of a non-linear stochastic system found in a much broader context than in fluid flows, elaborated to some extent in the theory of optimal control.
The practical implementation of a solution for such systems is a significant challenge. The common practice consists of a Monte Carlo approach that involves a computationally costly, slowly converging accumulation of the solution statistics. A recent method [1] avoids this difficulty by giving direct access to the statistical distribution of the solution. Largely inspired by control theory, this approach provides a useful tool that promises to facilitate the accounting of noisy measures and to allow a straightforward implementation of control algorithms. At present, a generalization accounting for the nonlinearities is still missing. Although the mathematical formulation is relatively straightforward, it requires validation coupled with a considerable effort, which involves techniques in model reduction. This effort will certainly be required in order to obtain a practically viable implementation relevant for large complex systems.
Integrative modeling of physical systems.
Understanding the relationship between molecular structure and function remains a grand challenge in the physical sciences and relies heavily on the use of theoretical models. A major goal in this domain is to bring theoretical models to the point of being quantitatively predictive. However, as the complexity of the systems being studied increases, it is necessary to bring the models closer to experimental observations and, one finds it necessary to include observations from experiment. Given the multidisciplinary nature of the physical and life sciences, an important approach that requires new development is an innovative integrated protocol that combines experiment and computations to facilitate the interpretation of physical phenomena, including structure/function characterization. Integrative approaches, in particular, in structural biology means combining data from different sources and levels of resolution, from the atom to the cell, providing a structural and dynamical description of a living cell. This requires integrative approaches that bring together proteomics and genomics data gathered over the last decade, so that a coherent, integrated picture of cellular, and eventually of multicellular organisms, emerges. Such description would provide unprecedented information on living organisms, with far reaching implications for human health, but also for the engineering of unicellular species with unique characteristics that can be exploited for such diverse endeavors as environmental detoxification and energy production. The development of such integrative approaches is a major scientific challenge that requires both new concepts and new technologies to bridge description scales that range from atomic to multicellular, and to integrate processes that cover dynamics scales from the femtosecond to the day. The success of integrative approaches will also depend on the ability to form teams of scientists with very different background, yet a common language. It will necessitate the delivery of experimental and computational technologies that are both highly sophisticated and « user-friendly » so that the advances can be disseminated to a large community of scientists.
Biomolecular simulations.
Biomolecular simulations and computational biology are rapidly evolving fields where progress is, in part, driven by a massive increase in experimental data and computing resources. Despite tremendous progress in the field, key challenges remain and are frequently named as some of the most important limitations of molecular modelling in, for example, pharmaceutical research. While knowledge of protein structure is a necessary step in understanding function, interactions and dynamics are key components required for a more complete understanding. While constant progress is made, the development of new computational methods that will allow one to address fundamental science is required. Methods that use, for example, experimental information in combination with simulation is currently a promising approach to determine structures of biomolecular complexes that cannot be solved experimentally at high resolution (i.e. large complexes). This requires the integration of information gathered by different methods, such as X-ray crystallography, NMR, EPR, electron microscopy, small angle X-ray scattering, calorimetry, surface plasmon resonance and mass spectrometry and optical spectroscopies with simulation. Similarly, our knowledge of the thermodynamics of intermolecular interactions is still limited and compromises our ability to predict whether proteins will interact and form stable complexes. Indeed, the interrelation of entropy and dynamics in favouring or disfavouring association is still hotly debated. Simulation also provides a way to represent the functional dynamics of proteins, which are not revealed by ordinary structure determination methods. This is vital in the study of allosteric proteins that are regulated by ligand binding to a site remote from the one that is primarily responsible for the molecular function, a key set of pharmaceutical targets. The main challenges here are in predicting the changes in structure and dynamics induced by the binding of an allosteric modulator, predicting allosteric binding sites, and quantifying an allosteric response and its therapeutical relevance.
The increase in computing resources is aligned with the technical developments apparent in new computer architectures. While recent advances in hardware and simulation software have significantly extended the accessible time scales, for many of today's simulation programs, full use of modern computer architectures is not generally achieved, leading sub-optimal simulation experiments. Algorithmic improvements to accelerate Molecular Dynamics are necessary from both the view of floating‐point operations to increased levels of parallelization on different platforms, including massively parallel, next‐generation supercomputers or highly distributed, non-homogenous computing.