Network Comparison

网络无处不在，一个典型的例子就是Facebook这样的社交网站，然而人们对它们还有很多不了解的地方。即使对于两个网络有多相似这个看似简单的问题，到目前为止也没有普遍接受的方法来回答这个问题。拟议的项目将解决这个问题，如何比较网络。网络比较是相当重要的，例如，当试图比较生物体在不同形式的压力下的蛋白质-蛋白质相互作用网络时，或者当试图比较携带某种疾病的人与未患病的人的蛋白质-蛋白质相互作用网络时。统计数据将适用于任何类型的网络，并且它们可以用于跟踪网络随时间的变化。虽然为网络比较构建这样的统计可能相对简单，但对于统计测试，必须了解其概率特性。要获得这样的理解，需要相当多的概率方面的专业知识，因为为了建立一个有用的渐近结果，必须解决许多问题。同样，必须理解新统计的统计特性。他们如何很好地分离由稍微不同的模型生成的模型？如何实施合适的测试程序？初步研究表明，我们的方法可以用于重建基于蛋白质-蛋白质相互作用网络的系统发育树。如果得到证实，那么这是第一个表明蛋白质-蛋白质相互作用网络的拓扑结构本身就包含有关进化的信息的结果。这一结果在生物学上引起了相当大的兴趣，其生物学意义必须加以详细研究。经验表明，网络能够吸引想象力，是外联活动的合适主题。因此，我们打算编写一些针对普通听众的讲座，首先通过部门联系以及在公共科学论坛上向学校提供。牛津火花门户网站是一个在线公共科学网站，将会有一个关于我们研究的博客。

Networks are ubiquitous, a prime example being social network sites such as Facebook, and yet there is a lot which is not understood about them. Even for the seemingly simple question of how similar two networks are, to date there is no generally accepted method available to answer this question. The proposed project will address exactly this issue, how to compare networks. Network comparison is of considerable importance for example when trying to compare protein-protein interaction networks of organisms under different form of stress, or when trying to compare protein-protein interaction networks of people carrying a certain disease to that of people without the disease. The statistics will be applicable to any types of networks, and they could be used to track changes in networks over time. While constructing such a statistic for network comparison may be relatively straightforward, for statistical tests its probabilistic properties have to be understood. Achieving such understanding requires considerable expertise in probability, as there are a number of questions which have to be tackled in order to establish a useful asymptotic result.Similarly, the statistical properties of the new statistics have to be understood. How well do they separate models which are generated from just slightly different models? How can suitable test procedures be implemented? Preliminary studies have shown that our method can be used to reconstruct phylogenetic trees based on protein-protein interaction networks. If confirmed, then this is the first result which shows that the topology of protein-protein interaction networks alone contains information about evolution. This result is of considerable interest in biology, and its biological implications have to be studied in detail. Experience has seen that networks are able to capture the imagination, and are a suitable topic for outreach activities. Hence we intend to develop some lectures aimed at a general audience, to be offered to schools, via departmental contacts in the first instance, as well as at public science forums.There will be a blog about our research on the Oxford Sparks portal, an on-line public science website.