Identifying Drug Repurposing Candidates for Rare Neuro-muscular Disorders, Using Different AI Methods on the Literature Knowledge Graph

This repository contains the code and resources accompanying the work:

Identifying Drug Repurposing Candidates for Rare Neuro-muscular Disorders Using Different AI Methods on the Literature Knowledge Graph (2025)

Papadimas, F., Svolou, S., Bougiatiotis, K., Aisopos, F., Krithara, A., and Paliouras, G.

This work was conducted in the context of the SIMPATHIC project, funded by the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No. 101080249).

Overview

Drug repurposing is a critical yet challenging task for rare diseases, where limited patient populations and sparse curated biomedical evidence hinder traditional drug discovery pipelines. This project presents a computational framework for drug–disease link prediction, integrating heterogeneous biomedical evidence into a unified literature-based knowledge graph.

Focusing on seven rare neurological, neurometabolic, and neuromuscular disorders:

1. SpinoCerebellar Ataxia type 3 (SCA3)

2. Congenital NeuroTransmitter defects (CNT)

3. Pyridoxine Dependent Epilepsy (PDE )

4. Congenital disorder glycosylation (PMM2)

5. Zellweger Spectrum Disorders (ZSD)

6. Myotonic Dystrophy type 1 (DM1)

7. Congenital Myasthenic Syndrome (CMS)

we construct a disease-centered biomedical knowledge graph and evaluate multiple artificial intelligence approaches.

Link Prediction Approaches

• Text-based similarity Baseline
• Rule-based inference (AnyBURL)
• Graph Neural Networks (R-GCN, CompGCN)
• Path-based methods (Path Analysis, PAM)

Disease similarity information derived from text embeddings is incorporated to enhance predictions.

Ensemble Prediction

Our ensemble model aggregates predictions from the best-performing models using a weighted ranking strategy, producing consolidated drug candidate lists for each disease.

Evaluation

• Quantitative evaluation using MRR and Hits@K

• Per-disease performance analysis

Expert manual validation of top-ranked drug candidates

Top-ranked drug candidates for each disorder were reviewed by domain experts using a structured annotation scheme (e.g., Investigated – does not work, Treats a symptom).