Network-Based Prioritisation Identifies More Candidates Beyond a Phenotype-Based Comparator in Alzheimer’s Disease: A NeDRex/REPO4EU Case Study

Abstract

Background Computational drug repurposing can prioritise candidate drugs through phenotype-based similarity or through network-based relationships between disease-associated genes and drug targets. Whether these approaches identify overlapping or distinct candidate spaces remains an important methodological question. Methods We compared DIAMOnD-based network prioritisation with an HPO→MedDRA→SIDER phenotype-based comparator in three disease modules—Alzheimer’s disease, pulmonary embolism, and ovarian cancer—using NeDRex/REPO4EU resources. We quantified the proportion of candidates assigned to symptom classes outside those captured by the phenotype-based comparator. Results Alzheimer’s disease was the only module with sufficient phenotype-derived candidates for meaningful statistical comparison (network-based: n=562, of which 316 candidates had symptom-class annotation; phenotype-based: n=288, of which 194 had symptom-class annotation). Among annotated candidates, DIAMOnD identified a significantly larger proportion of cross-symptom-class candidates than the phenotype-based comparator (253/316, 80.1% vs. 130/194, 67.0%; two-sided Fisher’s exact test, OR=1.98, 95% CI 1.32–2.97, p=0.0011). The two candidate sets showed limited overlap, sharing 68 of 782 total unique DrugBank identifiers (Jaccard index=0.087). For pulmonary embolism and ovarian cancer, sparse phenotype-derived candidate sets (n=0 and n=4, respectively) precluded comparable statistical inference. Conclusions In the evaluable Alzheimer’s disease case, network-based and phenotype-based prioritisation identified largely non-overlapping candidate sets, with limited overlap between the two approaches (Jaccard index=0.087). This finding does not establish mechanism-based disease reclassification, demonstrate that network proximity reflects a shared biological disease mechanism, or indicate two fully distinct, mutually exclusive candidate spaces. Rather, it supports the use of network-derived candidates as hypotheses for subsequent biological, translational, and disease-context-specific validation.

Authors 1

1. Ute Werner geb. Kröger corresponding Aachen

   Universitätsklinikum Aachen · Luisenkrankenhaus

   Affiliation as printed

   Luisenhospital

   Uniklinik Aachen