SUPR
Generating Knowledge Graph Embeddings for UMLS
Dnr:

NAISS 2024/22-1113

Type:

NAISS Small Compute

Principal Investigator:

Nicolas Pielawski

Affiliation:

Uppsala universitet

Start Date:

2024-09-01

End Date:

2025-04-01

Primary Classification:

10201: Computer Sciences

Webpage:

Allocation

Abstract

Deep Learning methods processing medical data may need to perform operations on patient-related information such as symptoms, diagnoses, and drugs called terms. Those terms are usually coded using various hierarchical systems (e.g., ICD-10, SNOMED, ATC, ...) which explain the relationships between the terms (e.g. [diagnosis A] [causes] [symptom B]). An AI trained on a dataset of patients could only learn about the relation between terms appearing in the dataset. During inference, the AI would not be able to generalize to new terms that it has not seen during training. This project aims at transforming terms into semantic embeddings by using knowledge graph embedding methods on the UMLS dataset, a graph-based knowledge dataset which contains most of the (known) relationships between those terms.