Section: Overall Objectives

Research axes

The goal of the NANO-D group is to help current and future designers of nanosystems, i.e. systems studied or designed at the atomic scale (whether natural or artificial, independently of the application domain, including structural biology, material science, chemistry, etc.) by developing the foundations of a software application which will run on a desktop computer, and will allow for efficient analysis, design, modeling and simulation of nanosystems.

To achieve this, we will be developing a series of adaptive methods and algorithms that allow users to focus computational resources on the parts of the models that they want to simulate, and that allow to finely trade between speed and precision.

In parallel, we will develop the architecture of a new desktop application for virtual prototyping of nanosystems, and will integrate all our algorithms into this application. Furthermore, the architecture of this platform will be open, so that independent developers may add modules, for multiple application domains (physics, biology, chemistry, materials, electronics, etc.). With this open platform, we will attempt to federate the research performed in computational nanoscience throughout the world.

This application is called SAMSON: “Software for Adaptive Modeling and Simulation Of Nanosystems”.

Our two research axes are:

1. Developing adaptive algorithms for simulating nanosystems

   • Defining adaptive Hamiltonians: In order to be able to perform simulations with good mathematical properties, we are expanding on our recent work on adaptively restrained Hamiltonians   [22], i.e. modified Hamiltonian representations of molecular systems that are able to switch degrees of freedom on and off during a simulation. These will allow us to finely trade between precision and computational performance, by choosing arbitrarily the number of degrees of freedom. Even though we have already obtained some promising results in this domain, our goal is to develop several different simplification methods.

   • Developing algorithms for incremental potential update: In order to benefit from performing adaptive particle simulations, we need to develop a series of algorithms that will take advantage of the fact that some (potentially relative) atomic positions are frozen. We have already demonstrated how this is possible for torsion-angle quasi-static simulation of classical bio-molecular force-fields [67], for neighbor search between large rigid molecules [21], and for bond-order reactive force-fields [25]. We are developing new algorithms for incremental neighbor search, energy and force updates corresponding to the adaptive Hamiltonians that we are defining.

2. Developing algorithms for modeling molecular interactions

   • Developing knowledge-driven methods, potentials and algorithms: Over time, more and more experimental information becomes available. One can use this information to predict and discover new types of molecular interactions and various mechanisms or molecular organization. For example, currently there are more than 50,000 protein structures of a high resolution stored in the Protein Data Bank [23] and over 500,000 structures of small molecules stored in the Cambridge Structural Database [17]. We are developing algorithms for protein-protein interactions and protein-ligand interactions.

   • Developing parametrization algorithms for interaction potentials: Molecular models typically require their own potential energy function (or a forcefield) to be assigned. However, the development of a new potential function is a very difficult and sometimes challenging task [43]. Therefore, we are developing algorithms for automatic parametrization of new potential functions for some particular representations of a molecular system.

   • Developing algorithms for exhaustive sampling: Some application domains, such as computational docking, cryo-EM rigid-body fitting, etc., require sampling in a low-dimensional space. For such applications it is advantageous to perform an exhaustive search rather than accelerated sampling [64]. Therefore, we are developing fast search methods to perform exhaustive search.