A two-tiered unsupervised clustering approach for drug repositioning through heterogeneous data integration

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery
• Network Analysis in Drug Repositioning

Background:

• Drug repositioning identifies new therapeutic uses for existing medications, reducing development risks.
• Computational methods analyze complex pharmacology networks to streamline drug repositioning.
• Clustering drugs simplifies large networks and serves as a basis for identifying repositioning candidates.

Purpose of the Study:

• To propose a two-tiered, unsupervised, drug-centric clustering approach for drug repositioning.
• To integrate heterogeneous drug data, including chemical, disease, gene, protein, and side-effect relationships.
• To compare the proposed method against existing data integration approaches using established metrics.

Main Methods:

• Employs a two-tiered clustering strategy: Growing Self Organizing Map (GSOM) for homogeneous profiles and graph clustering for drug-drug relations.
• Integrates diverse drug data profiles: chemical, disease, gene, protein, and side-effect relationships.
• Compares the approach against two existing heterogeneous data integration methods using Normalized Mutual Information (NMI) and Standardized Mutual Information (SMI).

Main Results:

• The proposed approach achieved superior NMI (0.66) and SMI (36.11) compared to existing methods (NMI: 0.60-0.64, SMI: 22.26-33.59).
• Successfully identified useful drug clusters for repositioning, unlike existing methods that struggled with graph clustering algorithms.
• Provided clinical evidence for four predicted repositioning candidates (Chlorthalidone, Indomethacin, Metformin, Thioridazine), validating the approach's reliability.

Conclusions:

• The two-tiered unsupervised clustering approach effectively clusters drugs and integrates heterogeneous data for reliable repositioning candidate identification.
• The method aligns with Anatomical Therapeutic Chemical (ATC) classification, enhancing its clinical relevance.
• Candidates identified consistently by multiple algorithms and with high confidence are more likely to be effective repositioning opportunities.