Abstract
To date, there are no effective treatments for most neurodegenerative diseases. Knowledge graphs can provide comprehensive and semantic representation for heterogeneous data, and have been successfully leveraged in many biomedical applications including drug repurposing. Our objective is to construct a knowledge graph from literature to study relations between Alzheimer's disease (AD) and chemicals, drugs and dietary supplements in order to identify opportunities to prevent or delay neurodegenerative progression. We collected biomedical annotations and extracted their relations using SemRep via SemMedDB. We used both a BERT-based classifier and rule-based methods during data preprocessing to exclude noise while preserving most AD-related semantic triples. The 1,672,110 filtered triples were used to train with knowledge graph completion algorithms (i.e., TransE, DistMult, and ComplEx) to predict candidates that might be helpful for AD treatment or prevention. Among three knowledge graph completion models, TransE outperformed the other two (MR = 13.45, Hits@1 = 0.306). We leveraged the time-slicing technique to further evaluate the prediction results. We found supporting evidence for most highly ranked candidates predicted by our model which indicates that our approach can inform reliable new knowledge. This paper shows that our graph mining model can predict reliable new relationships between AD and other entities (i.e., dietary supplements, chemicals, and drugs). The knowledge graph constructed can facilitate data-driven knowledge discoveries and the generation of novel hypotheses.
Introduction
Alzheimer’s Disease (AD) remains a global health crisis with no effective cure and a notoriously high failure rate for new drug development. Drug repurposing—finding new therapeutic uses for already approved drugs—offers a faster, cheaper, and safer alternative to traditional drug discovery. However, the sheer volume of biomedical literature is overwhelming; millions of papers contain "hidden" connections between chemicals and diseases that no human could possibly synthesize. This research addresses the challenge of mining this "dark data" to uncover non-obvious relationships between AD and various chemicals, drugs, and dietary supplements.
Methods
The researchers developed a multi-stage pipeline to transform unstructured text into actionable insights, as illustrated in the workflow below:
- Extraction: Over 113 million "semantic triples" (Subject-Predicate-Object, e.g., Aspirin-TREATS-Headache) were pulled from SemMedDB, a database derived from PubMed abstracts.
- Noise Reduction: To ensure data quality, rule-based filtering was applied to remove generic terms. Crucially, a PubMedBERT classifier was fine-tuned to verify that the extracted triples accurately reflected the relationships described in the original sentences.
- Knowledge Graph Construction: The final cleaned graph consisted of approximately 1.6 million triples involving 128,177 unique entities.
- Graph Embedding: Three algorithms—TransE, DistMult, and ComplEx—were tested to represent entities and relations in a continuous vector space. These models learn the structure of the graph to predict missing links (potential new treatments).
- Time-Slicing Evaluation: To simulate real-world discovery, models were trained on data published before 2019 and tested on their ability to predict "new" relationships that only appeared in literature published in 2019 and 2020.
Key Insights and Ideas
- The Power of Filtering: Raw literature mining is inherently noisy. By using a BERT-based classifier for "calibration," the reliability of the knowledge graph was significantly improved before the machine learning phase began.
- TransE Performance: Despite being a relatively simple "translational" model, TransE outperformed more complex bilinear models like DistMult and ComplEx for this specific AD dataset.
- Predicting the "Future": The model's ability to identify triples that hadn't been formally established by 2019 but were later supported by 2021 studies (such as the use of Amifostine for cognitive injury) demonstrates the predictive power of graph-based discovery.
- Holistic Repurposing: Unlike many studies focusing only on synthetic drugs, this research explicitly included dietary supplements (e.g., green tea, fiber, honey), recognizing that lifestyle and nutritional interventions are critical components of AD management.
Key Results
- Model Performance: As shown in the table above, TransE achieved the best results with a Mean Rank (MR) of 10.53 and a Hits@10 of 0.58, meaning the correct "hidden" relationship appeared in the model's top 10 recommendations 58% of the time.
- Top Candidates:
- Drugs: Identified Valacyclovir (an antiviral) and Tacrolimus (an immunosuppressant) as high-potential candidates. Recent trials have indeed begun investigating Valacyclovir's role in slowing AD progression.
- Chemicals: Amifostine and Betaine emerged as top-ranked candidates for AD prevention.
- Supplements: Dietary fiber and Green tea were highly ranked for their potential to reduce the risk of AD onset.
- Validation: For the top 10 predicted drug entities, supporting evidence was found in recent literature or clinical trials for 7 out of 10.
Conclusion
This study demonstrates that mining biomedical knowledge graphs can effectively predict reliable new relationships between AD and other entities. By integrating BERT-based calibration with graph embedding algorithms, the model successfully identified potential candidates for drug repurposing that were later validated by clinical evidence. This approach facilitates data-driven knowledge discovery and provides a robust framework for generating novel therapeutic hypotheses in neurodegenerative disease research.
Summary updated