Skip to main content

VAriable ResolutIon Algorithms for macroMOLecular Simulation

Periodic Reporting for period 2 - VARIAMOLS (VAriable ResolutIon Algorithms for macroMOLecular Simulation)

Reporting period: 2019-07-01 to 2020-12-31

The comprehension of the fundamental underpinnings of life, in particular at the molecular and macromolecular level, has immensely benefited, during the past few decades, of computer-aided methods. Most of these rely on techniques, such as molecular dynamics simulations, that enable the simulation of molecular systems, ranging from simple monoatomic liquids to complex molecules composed of millions of atoms; these “virtual experiments” enable researchers to investigate the behaviour of biologically relevant molecules, such as proteins and DNA, with a level of detail which is currently, or maybe fundamentally, inaccessible to experiment.
Computer simulations, in general, rely on the definition of a model of the system: this is a representation in terms of particles and interactions, such as atoms and the forces among them. Quite often, however, a fully-atomistic description of a large macromolecule can be either too intensive in terms of computing and storage resources, or too detailed and complex to analyse and comprehend, or both. Simpler models which describe the molecule at a lower resolution, the so-called coarse-grained models, offer a substantial advantage in that the reduced number of particles and the fewer interactions enable the simulation of large systems at a fraction of the computational cost of an equivalent all-atom simulation. However, a lower-resolution model naturally bears with it the disadvantages of a detail loss: this can be fatal for the reproduction and understanding of many phenomena of interest, which in turn depend on small-scale processes taking place in a small region of the molecule and reverberate up to the whole system.
The VARIAMOLS project aims to overcome this gap between expensive, overly-detailed all-atom models and efficient yet too blurred coarse-grained descriptions. This is achieved by means of a novel computer-based methods, a bottom-up modelling strategy in which the system learns during the simulation which parts can be simplified and to what extent this complexity reduction is allowed. In contrast to conventional multi-scale methods, the level of detail with which the macromolecule is modelled is not uniform across the system, but rather varies smoothly through the structure. This concurrent usage of various levels of resolution depending on the specific, local properties of the system optimises the balance between detail and efficiency. The method provides a deeper insight into how key biological macromolecules carry out their function, e.g. by means of concerted movements. On the long time scale, the outcome of this work will help to improve the development of antiviral and antibody-based drugs.
The first period of activity of the VARIAMOLS project has been extremely intense and productive. It has started with a relatively short setup phase, whose main objective consisted in the acquisition and installation of the computational infrastructure, and the selection of the other group members (two PhD students and two postdocs). During this period, two seminal papers have been published which conveyed the main message of the project (Diggins et al., JCTC 2019; Giulini and Potestio, Interafce Focus 2019); other two papers have been published whose work has been supported by the project (Tarenzi et al., JCTC 2019; Riccardi et al., Interface Focus 2019).
The main research lines of the project foreseen for this period have been started, with a particular focus on the algorithmic and methodological aspects. The main strategy for the construction of variable resolution models has been developed, discussed in detail, and implemented, and is currently in the validation phase. Additional strategies have been conceived and developed, some already in line with the project work plan, others coming out of the fruitful collaboration established with scientists outside of the group (e.g. M. Scott Shell, UCSB (USA), and Flavio Vella, Free University of Bozen (IT)).
Another relevant outcome of this first project period have been: the construction of a large dataset of proteins and their dynamics-based clustering for the identification of common structure-dynamics-function patterns; and the simulation of complex biological systems such as antibodies over time scales of at least 500 ns. These results provide the basis for the development of dynamics-based variable resolution modelling strategies capable of preserving crucial mechanical and dynamical properties of the reference model.
The results obtained insofar lay the basis for the automated, parameter-free construction of effective models of large biomolecules, proteins above all, which optimise the balance between accuracy and efficiency. Most importantly, the work carried will take advantage of dataset-based and machine-learning-based methods to boost the efficiency, however does not rely on them for the construction of the models. The approaches here developed have solid roots planted in physics and chemistry, and the gain in efficiency for practical application is always paired by an increase of comprehension of the fundamental properties of biomolecules. In contrast with available methods, the ones developed in the VARIAMOLS project aim at maximising understanding and efficiency at the same time, by providing researchers with tools that are not only effective and reliable for the investigation of complex biomolecules, but rather contribute themselves to comprehend the inner life of the latter.
Model optimisation flyer
Group photo