Childhood Cancer Data Initiative Annual Symposium (Abstract Registration): Submission #55

Submission information
Submission Number: 55
Submission ID: 150467
Submission UUID: 947014a0-f890-497d-a07c-efe8f2834db6

Created: Sun, 08/31/2025 - 14:07
Completed: Sun, 08/31/2025 - 14:52
Changed: Sun, 08/31/2025 - 14:52

Remote IP address: 10.208.28.30
Submitted by: Anonymous
Language: English

Is draft: No
Abstract Title: Graph Artificial Intelligence for Pediatric Oncology (GAIPO): a graph AI platform for precision oncology using pediatric clinical and omics data
Abstract: Advances in artificial intelligence (AI) are shifting the paradigm in precision medicine for pediatric cancer, including biomarker identification, drug discovery, and survival analysis. The Childhood Cancer Data Initiative (CCDI) ecosystem provides essential clinicogenomic data for deep learning in pediatric cancer research. To enable the efficient use of the CCDI resource for AI model training and implementation, we developed a generic graph AI platform, Graph Artificial Intelligence for Pediatric Oncology (GAIPO), as well as the standards of data, model, and pipeline, to streamline the use of CCDI clinicogenomics data of various data modalities from bulk and single-cell omics data to clinical information in AI development. GAIPO provides a comprehensive workflow: (1) Data fetching through CCDI Data Federation Resource API and cBioPortal API for clinical metadata, multi-omics, and spatial transcriptomics data; (2) Data modeling through mapping and harmonizing the fetched clinical and genomics data according to the CCDI data model; (3) Graph construction functionalities through GraphML from NetworkX; (4) Graph AI implementation of existing models; and (5) Post-analysis, such as vital feature selection and survival analysis. We demonstrate the capabilities of our GAIPO in treating two pediatric cancers, glioma and Wilms tumor, with potential applicability to other pediatric cancers. We apply graph AI models, MOGONET and PCGS, to integrate multi-omics data using explainable graph convolutional networks, allowing patient classification and biomarker identification. Notably, the new PCGS model identifies background-specific key features for biomarker discovery, risk group identification, and survival analysis in glioma and Wilms tumor, with potential applicability to other pediatric cancers.
Abstract:
Authors:
  1. First Name: Zanyu
    Last Name: Shi
    Degree(s): M.S.
    Organization: Department of Biostatistics & Health Data Science, Indiana University Fairbanks School of Public Health
  2. First Name: Aishwarya
    Last Name: Budhkar
    Degree(s): M.S.
    Organization: Department of Computer Science, Indiana University Luddy School of Informatics, Computing, and Engineering
  3. First Name: Waqas
    Last Name: Amin
    Degree(s): M.D.
    Organization: Indiana University School of Medicine
  4. First Name: Karen
    Middle Initial: E.
    Last Name: Pollok
    Degree(s): Ph.D.
    Organization: Indiana University School of Medicine, IU Simon Comprehensive Cancer Center
  5. First Name: Jing
    Last Name: Su
    Degree(s): Ph.D.
    Organization: Indiana University School of Medicine, IU Simon Comprehensive Cancer Center
  6. First Name: Kun
    Last Name: Huang
    Degree(s): Ph.D.
    Organization: Indiana University School of Medicine, IU Simon Comprehensive Cancer Center
Presenting Author: Zanyu Shi
Institution: Indiana University Indianapolis
Email Address: zanyshi@iu.edu