Measuring drug similarity using drug-drug interactions

Area of Science:

• Computational Biology
• Pharmacology
• Network Science

Background:

• Antimicrobial resistance necessitates innovative therapeutic strategies, including combination therapy.
• Predicting drug-drug interactions (DDIs) is crucial for effective combination therapy.
• Existing DDI prediction models often rely on drug chemical structures or mechanisms of action (MoA).

Purpose of the Study:

• To propose a novel drug similarity measure based on drug-drug interaction (DDI) network structure.
• To explore the application of this measure in unsupervised and semi-supervised learning for DDI prediction and functional annotation.
• To evaluate the performance of the proposed method against existing approaches.

Main Methods:

• Developed a drug similarity measure leveraging the topology of DDI networks.
• Integrated the similarity measure with unsupervised learning (clustering) for drug grouping and MoA inference.
• Utilized the similarity measure within semi-supervised learning frameworks to construct affinity matrices for predicting unknown DDIs.

Main Results:

• Unsupervised learning revealed drug groupings with similar MoA based on interaction patterns.
• Semi-supervised learning demonstrated the effectiveness of the similarity measure in predicting unknown DDIs.
• The proposed network-based similarity measure outperformed existing methods in DDI prediction tasks.

Conclusions:

• The proposed drug similarity measure based on network structure is effective and practical.
• This measure facilitates functional annotation of compounds with unknown MoA when combined with clustering.
• It enables accurate prediction of unknown DDIs when integrated with semi-supervised graph learning.