Скачать или смотреть dkNET Webinar: Building Knowledge Graphs Towards Transparent Biomedical AI 03/14/2025

Presenter: Jie Liu, PhD, Associate Professor, Computational Medicine and Bioinformatics, Computer Science and Engineering, University of Michigan

Abstract
Knowledge graphs have recently emerged as a powerful data structure to organize biomedical knowledge with explicit representation of nodes and edges. The knowledge representation is in a machine-learning ready format and supports explainable AI models. In this talk, I will describe several knowledge graphs built in my lab or in a larger team, including Genomic Knowledge Base (GenomicKB, https://gkb.dcmb.med.umich.edu/), Genomic Literature Knowledge Base (GLKB, https://glkb.dcmb.med.umich.edu/), and PanKgraph, the knowledge graph within the PanKbase project (https://pankgraph.org/). I will focus on the scope of these knowledge graphs as well as how they support transparent and explainable AI, including using them in machine learning tasks, reducing hallucination of large language models (LLM), and helping experimental scientists explore their data for scientific discoveries.

The top 3 key questions that PanKgraph can answer:
1. Which genes are highly expressed in specific tissues (e.g., pancreatic islets) and have known functional roles in T1D?

Explanation: The scRNAseq expression data allows researchers to filter genes that show high tissue specificity in T1D-relevant organs.

2. How do SNPs influence quantitative trait loci (QTLs) of specific genes (e.g., CFTR) in specific tissues (e.g., pancreatic islets)?

Explanation: This question helps in understanding how genetic variations regulate gene expression in disease-relevant tissues, leveraging QTL and fine-mapped eQTL data.

3. What are the most studied genes related to critical T1D-related pathways (e.g., TCR signal processing)?

Explanation: By integrating pathway annotations and literature references, PanKGraph can highlight top genes studied in T1D-relevant signaling pathways, helping researchers prioritize targets for further investigation.

Upcoming webinars schedule: https://dknet.org/about/webinar

Webinar host and organizer: Ko-Wei Lin, PhD. NIDDK Information Network (dkNET; https://dknet.org), [email protected]

Video edited by: Ko-Wei Lin