Knowledge graphs and their applications in drug discovery

Key Points

• The aim is to evaluate the limitations of knowledge graphs and graph-based machine learning in drug discovery.
• Assess existing link prediction algorithms in biomedical contexts.
• Analyze the biases present in the data and their impact on biological associations.
• Identify shortcomings in modeling causal relationships.
• Many link prediction algorithms do not effectively handle biases in biomedical data.
• Current models primarily highlight associations rather than causal dynamics.
• Significant challenges remain before fully leveraging knowledge graphs in drug discovery.