Repurposing non-pharmacological interventions for Alzheimer's disease through link prediction on biomedical literature

Yongkang Xiao, Yu Hou, Huixue Zhou, Gayo Diallo, Marcelo Fiszman, Julian Wolfson, Li Zhou, Halil Kilicoglu, You Chen, Chang Su, Hua Xu, William G. Mantyh, Rui ZhangScientific 2024 · DOI

TL;DR

This paper introduces a novel framework for repurposing non-pharmacological interventions (NPIs) for Alzheimer's Disease (AD) by building a comprehensive knowledge graph (KG) of NPIs and using link prediction models. The R-GCN model achieved the best performance, leading to the identification of 10 novel dietary supplements and 10 complementary and integrative health interventions with potential for AD prevention.

What Problem Does This Paper Address?

Alzheimer's Disease (AD) is a growing global health crisis with significant economic and social burdens. While pharmacological treatments exist, they often have side effects and limited efficacy. Non-pharmacological interventions (NPIs) like diet, exercise, and cognitive training show promise for improving cognitive function and potentially preventing AD, but their repurposing potential has been underexplored due to the sheer volume of biomedical literature and the difficulty in systematically identifying and validating new NPIs. This study aims to bridge this gap by developing a computational approach to discover and understand NPIs for AD prevention.

Methods

The core of this research is the creation of a biomedical knowledge graph (KG) called ADInt, which integrates information about NPIs from existing databases (SemMedDB and SuppKG). This KG was then used to train and evaluate various link prediction (LP) models. LP aims to infer missing relationships within a KG, essentially predicting potential new connections. The study compared four KG embedding models (TransE, RotatE, DistMult, ComplEX) and two graph convolutional network (GCN) models (R-GCN, CompGCN). After identifying the best-performing model (R-GCN), it was used to predict novel NPIs for AD prevention. Additionally, discovery patterns were employed to generate mechanistic pathways for these predicted NPIs, and real-world data (RWD) analysis was conducted to validate some of the findings.

Datasets

  • SemMedDB: A repository of semantic triples extracted from PubMed abstracts and titles, used for general biomedical knowledge.
  • SuppKG: A previously developed KG specifically focusing on dietary supplements (DS), created because the Unified Medical Language System (UMLS) lacked sufficient coverage of DS.
  • ClinicalTrials.gov: Used as a benchmark dataset to evaluate the LP models by identifying interventions used in AD clinical trials.
  • University of Minnesota (UMN) Clinical Data Repository: Real-world electronic health record (EHR) data from MCI patients, used for RWD analysis to validate predicted NPIs.

Key Insights and Ideas

  • KG-driven NPI Discovery: The study demonstrates the power of constructing a specialized KG (ADInt) to systematically mine the vast biomedical literature for potential NPIs for AD, moving beyond manual hypothesis generation.
  • R-GCN's Superiority: The Relational Graph Convolutional Network (R-GCN) emerged as the top-performing model for link prediction, highlighting its effectiveness in handling the complex, heterogeneous nature of the ADInt KG.
  • Mechanistic Pathways: The integration of "discovery patterns" to generate mechanistic pathways for predicted NPIs is crucial. It moves beyond simply identifying potential interventions to understanding how they might work, which is vital for clinical translation.
  • Validation through RWD: Using real-world EHR data to validate findings for interventions like psychotherapy and manual therapy provides strong evidence for their potential role in AD prevention, even if statistical significance wasn't reached in all analyses.

Key Results

  • ADInt KG: The constructed ADInt KG contains 162,212 nodes and 1,017,284 edges, representing a rich source of NPI-related information.
  • Model Performance: R-GCN achieved the best performance in both time-slicing evaluation (MR = 5.2054, Hits@10 = 0.8496) and clinical trials dataset evaluation (MR = 3.4996, Hits@10 = 0.9192).
  • Novel NPIs: Domain experts evaluated the top-scoring candidates, leading to the proposal of 10 novel dietary supplements and 10 complementary and integrative health (CIH) interventions for AD prevention.
  • Mechanistic Insights: Plausible mechanistic pathways were identified for photodynamic therapy and Choerospondias axillaris (a fruit).
  • RWD Validation: Psychotherapy and manual therapy techniques were validated using real-world data, showing a potential association with reduced risk of progression from Mild Cognitive Impairment (MCI) to ADRD, though statistical significance varied.

Limitations and Future Work

  • Limitations of PDT: The paper acknowledges that photodynamic therapy (PDT), while showing potential, has significant practical limitations for treating AD due to the difficulty of delivering light to the entire brain.
  • Statistical Significance in RWD: While RWD analysis showed promising trends for psychotherapy and manual therapy, the associations did not always reach conventional statistical significance, suggesting the need for larger studies.
  • Observational Nature of RWD: The RWD analysis is observational, meaning it cannot definitively rule out unmeasured confounding factors.
  • Future KG Integration: The authors suggest integrating ADInt with other comprehensive biomedical databases (e.g., DrugBank, KEGG) to expand the scope of analysis and discover more interventions.
  • Exploring Other LP Methods: Investigating link prediction methods based on language models is proposed as a future direction.
  • Expert Evaluation Scale: Currently, only the top 10 candidates from each scoring table are evaluated by experts due to resource constraints. Future work could involve evaluating larger sets of candidates.

Summary updated

Abstract

Non-pharmaceutical interventions (NPI) have great potential to improve cognitive function but limited investigation to discover NPI repurposing for Alzheimer's Disease (AD). This is the first study to develop an innovative framework to extract and represent NPI information from biomedical literature in a knowledge graph (KG), and train link prediction models to repurpose novel NPIs for AD prevention. We constructed a comprehensive KG, called ADInt, by extracting NPI information from biomedical literature. We used the previously-created SuppKG and NPI lexicon to identify NPI entities. Four KG embedding models (i.e., TransE, RotatE, DistMult and ComplEX) and two novel graph convolutional network models (i.e., R-GCN and CompGCN) were trained and compared to learn the representation of ADInt. Models were evaluated and compared on two test sets (time slice and clinical trial ground truth) and the best performing model was used to predict novel NPIs for AD. Discovery patterns were applied to generate mechanistic pathways for high scoring candidates. The ADInt has 162,212 nodes and 1,017,284 edges. R-GCN performed best in time slice (MR = 5.2054, Hits@10 = 0.8496) and clinical trial ground truth (MR = 3.4996, Hits@10 = 0.9192) test sets. After evaluation by domain experts, 10 novel dietary supplements and 10 complementary and integrative health were proposed from the score table calculated by R-GCN. Among proposed novel NPIs, we found plausible mechanistic pathways for photodynamic therapy and Choerospondias axillaris to prevent AD, and validated psychotherapy and manual therapy techniques using real-world data analysis. The proposed framework shows potential for discovering new NPIs for AD prevention and understanding their mechanistic pathways.
BibTeX
@inproceedings{xiao2024repurposing,
  author = {Yongkang Xiao and Yu Hou and Huixue Zhou and Gayo Diallo and Marcelo Fiszman and Julian Wolfson and Li Zhou and Halil Kilicoglu and You Chen and Chang Su and Hua Xu and William G. Mantyh and Rui Zhang},
  title = {{Repurposing non-pharmacological interventions for Alzheimer's disease through link prediction on biomedical literature}},
  booktitle = {Scientific},
  year = {2024},
  doi = {10.1038/s41598-024-58604-8},
}
Figure