REU: GDA at GSU - Projects

Research Projects

Students will participate in specific research projects related to graph data analytics within the overall cohort research experience. They will be introduced to the fundamentals of graph data analytics and work on real-world projects. They will get some hands-on experience with Python and MATLAB libraries on many machine learning and data mining models. They will generate a hypothesis, design and implement a research plan to test their hypothesis, and analyze the results with the help of research mentors. The students engaging in any of the tasks will be given co-authorship in the research papers generated. We present brief descriptions of various projects below.

Project 1: Graph Machine Learning and Application to Biomedical Problems

Faculty Mentor: Dr. Esra Akbas

Graph machine learning including Graph Neural Networks (GNNs), works on graph data structures by extracting the features of nodes and capturing complex dependencies in underlying systems. However, real-world networks' size and complexity present challenges for these models to apply to real-world problems including biomedical problems. Students will address this by developing structure-aware graph processing methods that preserve local and global connectivity while breaking noisy connections. They will explore critical topological properties, develop structure-aware compression methods, and design tailored architectures of GNN on compressed graphs. They will also apply developed models to sequence classification and clustering, drug repurposing, and protein function prediction problems. The project involves Python programming and tools such as NetworkX, Pytorch Geometric (PyG), DGL, and GraphNets, exposing students to graph mining, graph machine learning, and real-world problem applications.

Project 2: Influence Maximization via Diffusion on Graphs

Faculty Mentor: Dr. Mehmet Aktas

In this project, students will delve into the critical concept of modeling diffusion on graphs, a key aspect of graph data analytics with broad applications. The project emphasizes the significance of diffusion models in domains like social networks and transportation networks, where insights into information propagation and traffic congestion can inform strategies for rumor control and urban planning. The focus extends to the influence maximization problem on graphs, involving the identification of influential elements in a network that significantly impact its structure and functions. Students will develop various diffusion models, applying them to diverse graph types, including directed graphs and hypergraphs. The practical application of these methods will be explored in the realms of social influencer marketing and expertise detection, showcasing the real-world implications of graph diffusion models in business analytics research.

Project 3: Graph-Based Modeling of Viral and Microbial Transmission Dynamics

Faculty Mentor: Dr. Murray Patterson

When a contagious viral or microbial infection is first introduced to a location, it spreads locally between neighboring locations. However, distant locations can be even more well-connected than neighboring ones, e.g., via students who attend the same school, or cities that have multiple flights a day between them. While neighboring locations can be explained by physical (or Euclidean) distance, these more distant relationships---which seem to defy or distort the Euclidean metric---are better explained with a graph. Taking both of these into account is important to properly modeling transmission dynamics and predicting hotpots of viral or microbial infections such as Influenza, COVID-19, or HIV. In this project, students will analyze the surveillance data (genomic/proteomic sequencing, etc.) of such infections and build models of spread that can be used to prevent or mitigate outbreaks. Such a project involves sequencing data preprocessing and analysis, using bioinformatics tools such as samtools and bwa-mem, as well as programming in Python, using libraries such as PySAM and NetworkX.

Project 4: Leveraging Heterogeneous Graphs for Biomedical Data Analysis

Faculty Mentor: Dr. Farhan Tanvir

In this project, students will engage in a comprehensive biomedical project, delving into genomics, proteomics, and precision medicine. Throughout the project, students will develop innovative methodologies utilizing heterogeneous graphs and heterogeneous graph neural networks (HGNNs) to tackle critical issues, including gene-disease associations, protein function predictions, and patient-specific treatment recommendations. The focus starts with predicting gene-disease associations crucial for understanding disease genetics, followed by protein function prediction vital for drug discovery. The project culminates in the development of a recommendation system for patient-specific treatment plans in precision medicine, integrating diverse biological entities and interactions. Students will employ multi-modal learning, transfer learning, and generative AI techniques, addressing challenges like data integration and biologically plausible data generation. The project aims to advance representation learning methods, rigorously evaluating their impact on various bioinformatics tasks using publicly available datasets, contributing to the progress of bioinformatics and biomedical research.