Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150935

150935 - 3d-4-Big Data: 3d Printed Electronics Enabled 4d Visualization of Big Data Networks 

Current statistical and computational advances in network analysis lack the ability to understand the interconnectedness between network nodes in a tactile manner. Effective visualization of network graphs can not only identify underlying structure in the data, but are also critical for dimension reduction and training. The overall research goal is to design, develop, demonstrate, and evaluate a novel data visualization method that brings together data analytics, augmented reality (AR), and 3D Printed Electronics (3DPE - embedded electronics within 3D printed network nodes and LED strips at the edges). This novel approach is demonstrated for genetic associations between neuroimaging phenotypes and diseases in electronic health records (EHR) across the phenome. Our approach can yield novel image derived phenotype (IDP) biomarkers for complex diseases as well as provide pleiotropic gene targets for drug repurposing. We have established: (1) an open source Arduino toolkit to drive the 3DPE nodes and data network using open source low cost microcontroller boards (ESP32 WROOM DevKit), and (2) visualization of node labels and contextual information in mixed reality through AR lens modalities (e.g., Unity, Microsoft HoloLens) with the physical model to visualize 3D network models. We apply this to summary statistics from a transcriptome-wide association study on 2,124 IDPs (structural and diffusion MRI measuring cortical volume, cortical thickness, brain volumes, cortical grey/white contrast, etc.) from UK Biobank (UKB) on 38K samples of European ancestry across 13 brain tissues from GTEx v8 followed by gene-based colocalization. We mapped the significant genes to finemapped eQTL from PredictDB (using multivariate adaptive shrinkage models). Finally, we conducted an IDP guided phenome-wide association study on 12,494 fine-mapped eQTL across 664 ICD-10 codes (n=452,595 samples). We found that 3D-4-Big Data can seamlessly create mixed reality 3D network models for GWAS and TWAS studies with preprocessed high dimensional data and is effective in mapping network models with multiple modalities. For example, the use of AR for the labeling of genes and drug pathways; or the use of embedded electronics for dynamic networking (flowing lights on LED strips to identify edge direction, LED intensity to identify strength of relationship). We have developed a novel immersive 3DPE enabled 4D visualization of big data networks to better understand the complexity of high density data networks in human genetics. This study demonstrates the utility of this approach as a powerful tool for interactively visualizing pre-processed high dimensional multilayered data (e.g., genomics, transcriptomics, EHR) for the purposes of statistics education.

Presenting Author: Julian Kim Pennsylvania State University

Presenting Author Biography: Julian Kim is a 3rd year undergraduate student at the Pennsylvania State University, studying Mechanical Engineering with minors in Math and Engineering Leadership Development. He is a member of the Schreyer Honors College and the Millennium Scholars Program. He works as an undergraduate research assistant at the Systems for Hybrid-Additive Process Engineering (SHAPE) Lab under faculty advisor Dr. Guha Manogharan. Outside of classes, Julian is a member of Penn State's Club Climbing Team and is the principal oboist for the Campus Orchestra. He is also the president of Penn State's Global Brigades Water and Sanitation Health (GB WASH) Chapter.

Authors:

Julian Kim Pennsylvania State University
Clayton Colson Pennsylvania State University
Alex Fatemi Pennsylvania State University
Patrick Dudas Pennsylvania State University
Guha Manogharan Pennsylvania State University
Yogasudha Veturi Pennsylvania State University