Systematic discovery of disease-modifying targets by prediction from knowledge graph-based AI model and experimental validation: Parkinson's disease case.

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

The development of disease-modifying therapies (DMTs) for Parkinson's disease (PD) remains a critical unmet need. Despite extensive research efforts, no therapy capable of slowing or halting PD progression has been approved. Here, we apply a knowledge graph-based artificial intelligence (AI) framework, combined with subgraph-level enrichment-based re-prioritization, to identify novel PD-modifying targets without requiring disease-specific training or additional experimental datasets. Using model-derived PD association scores, we obtained 2527 predicted targets. To evaluate their connectivity to an expert-curated set of PD-associated genes, we performed subgraph-level over-representation analysis and identified 74 targets whose local subgraphs were significantly enriched for PD-relevant context. After applying novelty filters, five candidates remained, among which tripeptidyl peptidase 1 (TPP1) emerged as a compelling PD DMT target. The predicted association among PD, α-synuclein, and TPP1 within the subgraph was supported by differential expression analyses of publicly available RNA-seq datasets and validated experimentally in a human cell-based α-synuclein aggregation model. TPP1 expression was elevated in neuromelanin-positive dopaminergic neurons in late-stage PD, and its knockdown increased α-synuclein aggregation, suggesting a protective role in α-synuclein homeostasis. Structural modeling of AlphaFold-Multimer further revealed a substrate-like interface between α-synuclein and the TPP1 catalytic triad, consistent with a potential proteolytic mechanism of α-synuclein clearance. Together, these findings identify TPP1 as a previously underappreciated and mechanistically plausible PD DMT target and demonstrate how static knowledge graphs can be transformed into interpretable, disease-focused target discovery systems. By integrating explainable subgraph structures with enrichment-based re-prioritization, this framework provides a generalizable strategy for therapeutic target identification across indications.

Related topics