Learning disease relationships from clinical drug trials

Area of Science:

• Computational biology
• Medical informatics
• Translational medicine

Background:

• The assumption that more granular data always leads to better medical learning is being challenged.
• Clinical trial metadata may contain untapped expert knowledge regarding disease biology.

Purpose of the Study:

• To test the hypothesis that latent expert knowledge within clinical trial metadata can be leveraged to understand disease biology.
• To demonstrate that information about underlying disease mechanisms can be extracted from the collective body of clinical trials.

Main Methods:

• Extracted free-text metadata from 93,654 clinical drug trials.
• Developed a novel representation for comparing trials and constructing a disease network based on metadata.
• Interpreted each trial as a synthesis of expert knowledge on biological mechanisms and disease-drug relationships.

Main Results:

• A novel disease network was constructed using only clinical trial metadata.
• The generated network demonstrated significant agreement with networks derived from genetic data and the Medical Subject Headings (MeSH) taxonomy.
• The network also revealed unique disease similarities not found in existing classifications.

Conclusions:

• The findings support the hypothesis that latent expert knowledge exists within clinical trial metadata.
• Unique disease relationships identified in the network can inform future research, drug repurposing, and drug design.
• This study exemplifies the use of human experimental data to generate biologically relevant insights and proposes new strategies for connecting clinical outcomes to biological research.