Studying the Properties of Cellular Materials with GPU Acceleration

Abstract

There has always been a great interest in cellular behaviour. From the molecular level, studying the chemistry of the reactions that occur in cell, and the physical interactions between those molecules, to the scale of the cell itself and its behaviour in response to various phenomena. Suffice it to say, that cellular behaviour is highly complex and, therefore, it is difficult to predict how cells will behave or to even describe their behaviour in detail. Traditionally cell biology has been done solely in the laboratory. That has always yielded interesting results and science. There are some aspects of phenomena that, due to cost, time, or other factors, need to be studied computationally. Especially if these stimuli occur on very short or long time scales. Therefore, a number of models have been proposed in order to study cell behaviour.

Unfortunately, these methods can only be used in certain situations and circumstances. These methods can, and do, produce interesting and valid results. Yet there is not really any model available that can be used to model more than one or two kinds of cell behaviour. For example, methods that can show cell sorting do not necessarily show packing. Furthermore, many of the models in the literature represent cells as collections of points, or polygons, so cellular interactions at interfaces cannot be studied efficiently. The goal of the work presented here was to develop a three dimensional model of cells using Molecular Dynamics. Cells are represented as spherical meshes of mass points. And these mass points are placed in a force field that emulates cellular interactions such as adhesion, repulsion, and friction. The results of this work indicates that the model developed can reproduce qualitatively valid cellular behaviour. And the model can be extended to include other effects.

It must also be recognized that \gls{md} is very expensive computationally. Especially in the case of this model as many mass points are needed in the cellular mesh to ensure adequate spatial resolution. Higher performance is always needed either to study larger systems or to iterate on smaller systems more quickly. The most obvious way to alleviate this problem is too use high performance hardware. It will be shown that this performance is most accessible, after some effort, with \gls{gpu} acceleration. The model developed in this work will be implemented with \gls{gpu} acceleration. The code generated in this way is quite fast.