Effective holistic characterization of small molecule effects using heterogeneous biological networks

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Clinical trial attrition due to efficacy and safety concerns necessitates improved drug development methods.
• Existing drug discovery platforms often lack comprehensive integration of diverse biological data.

Purpose of the Study:

• To enhance the Computational Analysis of Novel Drug Opportunities (CANDO) platform for accurate therapeutic candidate generation.
• To integrate heterogeneous biological data for a multiscale understanding of drug behavior.
• To leverage indirect data, such as side effect profiles, for improved drug-indication association prediction.

Main Methods:

• Integrated diverse datasets including drug side effects, protein pathways, protein-protein interactions, and Gene Ontology into a human interactome network.
• Developed a 'multiscale interactomic signature' for each compound, representing its functional behavior as real-valued vectors.
• Utilized a random forest machine learning model trained on compound-protein interaction scores to predict drug-indication associations.

Main Results:

• Demonstrated significant biological information capture within integrated networks, particularly from side effect data, enhancing platform performance.
• Successfully generated novel drug candidates for colon cancer and migraine disorders, validated by literature.
• Highlighted potential applications in predicting drug associations for mental disorders and cancer metastasis.

Conclusions:

• The enhanced CANDO platform accurately relates drugs in a multitarget, multiscale context.
• The integration of indirect data, like side effect profiles and pathway information, is crucial for generating accurate putative drug candidates.
• This interactomic pipeline offers a powerful approach for drug discovery, repurposing, and design.