Genes produce proteins that interact with each other to carry out cellular functioning. Thus, the field of network biology, which deals with inference and analysis of protein-protein interaction networks and other types of biological networks, has a potential to revolutionize our understanding of cellular functioning, evolution, disease, and therapeutics. Traditional biological network research has dealt with static representations of biological -omics data, even though cellular processes are dynamic. To address this gap, as its intellectual merit, this project introduced efficient and accurate computational approaches (or algorithms) for inference and analysis of dynamic biological networks. As an example of the project's broader impact, the new approaches were evaluated in an application of studying human aging as a representative dynamic biological process of importance to global health, because aging is known to associate with many prevalent diseases. Because network research spans many domains, open-source software implementing the new approaches was provided as part of their publications, and also, the new approaches were evaluated on other types of real-world networks. A prominent example of a non-biological application under this project was evaluating the predictive power of dynamic social interactions on people's mental health.

In more detail, key novelties of this project regarding inference of a dynamic biological network are follows. The aging process is highly influenced by genetic factors. Hence, it is important to identify human aging-related genes. Doing this via wet lab experiments is hard because of ethical constraints and long human life span. So, computational identification (i.e., prediction) of aging-related genes has received significant attention. Gene expression-based methods for this purpose do account for aging-specific information (which genes are "active" at which age) but not for interactions between genes' protein products. On the other hand, protein-protein interaction (PPI) network-based methods for this purpose account for these interactions, but current PPI network data are context-unspecific, spanning different biological conditions (in this project, different ages). Instead, this project dealt with integration of aging-specific gene expression data and context-unspecific PPI network data into an aging-specific PPI subnetwork. Then, it demonstrated through a series of studies and methodological innovations for inference of such a subnetwork (first static then dynamic subnetwork, first unweighted then weighted subnetwork, first using older then newer gene expression and PPI data, etc.) that analyzing a weighted dynamic aging-specific subnetwork yields higher accuracy when predicting aging-related genes than using a weighted or static aging-specific subnetwork as well as the entire context-unspecific PPI network. So, this project's computational approaches for inference of a weighted dynamic aging-specific subnetwork could guide with higher confidence than the existing approaches the discovery of novel aging-related gene candidates for future wet lab validation. Importantly, beyond "just" aging, the approaches are generalizable to any dynamic biological process, including disease progression.

Key novelties of this project regarding analysis of an (inferred or existing, biological or non-biological) dynamic network are as follows. The idea of graphlets (small subgraphs, basic building blocks of real-world networks), which had been well established in static biological network research, was extended to allow for graphlet-based analyses of dynamic biological networks. The dynamic graphlets were then used in several downstream computational tasks, including clustering of a dynamic network (as opposed to traditional static network clustering). Specifically, two types of clustering approaches were developed. One questioned the traditional assumption that it is densely interconnected nodes in a dynamic network that should form a cluster (or functional module) and instead argued that it is topologically similar nodes that should be clustered together. The other approach argued that traditionally the problem of dynamic network clustering (or community detection) is approached by assuming either that each time point has a distinct community organization or that all time points share a single community organization, but that the reality likely lies between these two extremes. To find the compromise, this approach simultaneously partitions a dynamic network into contiguous time segments with consistent community organization and finds this community organization for each segment. Additional methodological novelties were proposed when predicting human aging-related genes, including new approaches for supervised node classification in a weighted dynamic network, as traditional approaches for this task can handle only unweighted or static networks.

As another example of this project's broader impact, research supervision and career mentoring were offered to 7 graduate students, 11 undergraduate students, and a high school student. Of the 19 students, 47% are women. Community outreach was organized to local middle school girls and high school students. Namely, workshops on network research were offered at the "Expanding Your Horizons in Science and Mathematics" career conference for middle school girls annually since 2015 and until the pandemic hit in 2020, with about 40 girls participating each year. Also, a collaboration was carried out with a local biology high school teacher to actively engage with their students with the goal of attracting them to the field of computing.

Last Modified: 06/20/2022
Modified by: Tijana Milenkovic