Automating Mechanism Discovery in Parkinson’s Disease: An NLP-Based Knowledge Graph Expansion Framework

In this study, we present a automated framework that leverages natural language processing (NLP) techniques to extract, encode, and integrate biological knowledge into a Parkinson’s Disease (PD)-focused mechanism knowledge graph (KG). To address the challenge of keeping biomedical KGs current with the accelerating pace of scientific discovery, we developed an end-to-end pipeline combining named entity recognition, relation extraction, and Biological Expression Language (BEL) generation. Using abstracts from recent literature retrieved via PubMed APIs, we applied tools such as SciSpacy, BioBERT, and OpenNRE to identify causal relationships relevant to PD pathophysiology. These relationships were translated into BEL triples and curated by domain experts for quality assurance. The validated triples were then incorporated into a PD-specific KG using PyBEL, with integration of additional mechanistic data from KEGG and Reactome. Analysis of the updated graph using NetworkX and BiKMi identified novel upstream regulators of α- synuclein phosphorylation, including druggable targets such as PINK1 and SIRT2. Our results demonstrate the feasibility and utility of using an NLP-guided approach for mechanism discovery and knowledge graph enrichment in neurodegenerative disease research.