Computing Drug-Drug Similarity from Patient-Centric Data

Area of Science:

• Pharmacology
• Computational Biology
• Medical Informatics

Background:

• Drug-drug similarity is crucial for pharmaceutical development, typically relying on drug-centric data.
• Existing methods use side effects, interactions, targets, and chemical structures to assess drug similarity.
• A gap exists in leveraging patient-reported outcomes for drug similarity analysis.

Purpose of the Study:

• To propose and evaluate a novel computational method for measuring drug-drug similarity.
• To explore patient-centric data from social media as a source for drug similarity computation.
• To assess the feasibility and accuracy of using patient narratives for drug similarity.

Main Methods:

• Extracted patient reviews of anti-epileptic drugs from online healthcare communities.
• Preprocessed text data using Natural Language Processing (NLP) techniques.
• Applied four text similarity methods to compute drug-drug similarities and generated ranking matrices.
• Validated results against ground-truth similarities from DrugSimDB using Pearson correlation.

Main Results:

• Demonstrated the feasibility of using patient-centric social media data for drug-drug similarity.
• Identified text similarity methods suitable for analyzing patient narratives.
• Generated drug-drug similarity matrices based on patient experiences.
• Showcased potential for novel drug discovery and development insights.

Conclusions:

• Patient-centric data from social media represents a viable and novel source for drug-drug similarity assessment.
• This approach complements traditional drug-centric methods, offering new perspectives.
• Findings support the integration of real-world patient data in pharmaceutical research.