EAGER: A Novel Algorithmic Framework for Discovering Subnetworks from Big Biological Data

大数据分析通常追求的目标之一是从数据中找到有趣的模式。当数据很大并且从底层网络集合中收集时，例如分子谱数据，推断分子子网络成为从生物大数据中发现知识的一个有前途的解决方案。如何有效地利用数据中海量的异构信息是阻碍发现的主要障碍，例如如何整合数据矩阵的行和列信息，以有效地探索可能子网的复杂空间。最近的一项研究（由PI和其他人）导致了新的算法被引入到这个领域。不幸的是，这些算法中的大多数既不是专门为生物大数据设计的，也不能很好地处理生物大数据。该项目的主要目标是开发量身定制的算法和软件工具，以便从不断增长的生物大数据中更好地发现子网。这个项目更广泛的意义和重要性主要体现在三个方面。首先，本提案中开发的子网算法和软件工具将对许多通常需要子网结构的科学领域具有广泛的适用性；这包括生物、计算、医学和社会科学等学科。建立一个有效和用户友好的软件工具箱将进一步为培训和教育这些科学领域的学生提供丰富的资源，从而有助于确保国家学术竞争力。第二，定期举办的外展活动将为教育各级学生和广大社区提供创新的学习模式。最后，代表性不足的群体，如女性和少数民族学生，将通过有针对性的征聘和信息传播参与其中。技术上，将开发和实现一种新的算法框架，即子网，从同一生物大数据中提取的分子丰度值和共调控分子集中共同发现子网。前者对应于数据矩阵的行，后者对应于数据矩阵的列。以前的研究只关注列或行，而不是同时关注两者。将引入一种新的多标准评分和搜索范式，并开发和实施一种新的子网算法，以有效、可靠地从生物大数据中提取底层子网。这些技术具有变革性，因为它们适用于大数据由底层网络“发射”的许多其他科学领域。算法和工具将在使用标准措施的模拟数据集上进行系统评估。

One of the commonly pursued objectives in big data analytics is to find interesting patterns from data. When the data is big and collected from an ensemble of underlying networks, such as molecular profiling data, inferring molecular subnetworks emerged as a promising solution to knowledge discovery from biological big data. A main barrier impeding the discovery is how to effectively use the massive and heterogeneous information from the data, e.g., how to integrate information from rows and columns of the data matrix to efficiently explore the complex space of possible subnetworks. A recent line of research (by the PI and others) has resulted in new algorithms being introduced to this area. Unfortunately, most of these algorithms are neither specifically designed for nor work well with biological big data. The main goal of this project is to develop tailor-made algorithms and software tools to obtain better discovery of subnetworks from ever-increasing biological big data. The broader significance and importance of this project fall into three main areas. First, the subnet algorithms and software tools developed in this proposal will have broad applicability for many scientific domains wherein subnetwork structures are usually desired; this encompasses disciplines ranging from biological, computational, medical and social sciences. The creation of an efficient and user-friendly software toolbox would further provide rich resources for training and educating students in these scientific domains, thereby helping to ensure national academic competitiveness. Second, the regularly scheduled outreach activities will provide an innovative learning model for educating students of all levels and the community at large. Finally, the under-represented groups, such as female and minority students, will be involved through targeted recruiting and information dissemination.Technically, a novel algorithmic framework, i.e., subnet, will be developed and implemented to discover subnetworks jointly from molecule abundance values and co-regulated molecule sets extracted from the same biological big data. The former correspond to the rows and the latter correspond to the column of the data matrix. Previous research has focused on either columns or rows but not on both simultaneously. A novel multi-criteria score-and-search paradigm will be introduced and a novel subnet algorithm will be developed and implemented to efficiently and reliably extract underlying subnetworks from biological big data. These techniques are transformative in that they are applicable to many other scientific areas where big data are "emitted" by the underlying networks. The algorithms and tools will be systematically evaluated on simulation data sets using standard measures.