A machine-learned computational functional genomics-based approach to drug classification.

Area of Science:

• Computational pharmacology
• Genomics
• Data science

Background:

• Big data in pharmacology enables novel data science approaches.
• Linking drugs to pathophysiologic processes offers a phenotypic path for drug discovery and repurposing.
• This study compares functional genomics criteria with traditional drug target-based classification.

Purpose of the Study:

• To compare the performance of a functional genomics-based criterion against the traditional drug target-based classification for drug categorization.
• To evaluate the utility of data science approaches in pharmacology for drug discovery and repurposing.

Main Methods:

• Constructed "drug target versus biological process" matrices from DrugBank and Gene Ontology databases.
• Utilized self-organizing maps (SOM) to project matrices onto a 50x82 artificial neuron grid.
• Visualized high-dimensional feature space using a U-matrix to identify cluster structures.

Main Results:

• The U-matrix successfully classified classical analgesics into opioid and non-opioid classes using both methods.
• Functional genomics approach inherently captured drugs' modulatory effects on biological processes.
• Identified known pharmacological actions, such as lipid signaling for NSAIDs and neuronal transmission for opioids.

Conclusions:

• Functional genomics-based criteria are as suitable as traditional target-based criteria for computational drug classification.
• Supports the use of functional genomics in computational system pharmacology for drug discovery and repurposing.