LJKDeterministic Models and Algorithms: EDPMOISE Seminar

On Thursday November 6 2008 at 11h00 in Salle 1  Tour IRMA

Seminary of Mr Stéphane REDON (INRIA RhoneAlpes)

Adaptive Molecular Dynamics

Summary

According to some estimates, the world market for nanotechnologyrelated products and services, including both traditional (e.g., drug, chemical,
material design and electronics) and novel nanotechnologies, will reach one trillion dollars by 2015.
Similar to what happened in the twentieth century, where progress in computeraided engineering methods largely contributed to the development of
macroscopic objects such as cars, planes, and many other manufactured objects, computing will likely play an essential role in the development of
nanotechnology.
Computeraided design of a nanosystem is still an extremely challenging problem, however, due in particular to the complexity of the underlying
physics. To address this problem, it is usually tempting to increase the computational resources (which may be expensive), or simplify the models
(which may become too simple to be useful).
Frequently, these simplification methods involve representations in reduced coordinates (e.g. modeling the molecule as an articulated body), where subsets of atoms are replaced by idealized structures, or performing normalmode or principal components analysis in order to determine the
essential dynamics of the system. However, current geometry or dynamics simplification methods have a fundamental flaw: they are unable to automatically determine the level of detail which best describes a given molecular interaction. Thus, the scientist must have some prior structural knowledge about the interaction to be modeled before choosing the best
representation; he or she must choose the simplest representation of the
molecules, i.e. the most efficient in terms of computational cost, which still allows precise simulations of the biological phenomenon under study.
This talk will present our current work on designing a unified theoretical
framework for Adaptive Molecular Dynamics (AMD). Unlike previous "a priori"
simplification approaches, adaptive molecular dynamics can predict the set
of active joints at each time step, based on the current state of the
molecular system (atom positions and velocities), the internal forces (van
der Waals, electrostatic, dihedral), the applied external forces (e.g., the
force applied by the user through a haptic device), and the precision
threshold specified by the user (i.e. maximum number of active joints, or
total acceleration threshold). We will introduce some underlying algorithms
and present results obtained with our interactive molecular modeler on
biological examples. We will also discuss further applications, including rapid prototyping of molecular dynamics simulation, nanosystem design, study of macromolecular motions and docking, de novo protein design.
