Quantifying the relationships among drug classes.

Area of Science:

• Drug discovery and development
• Chemoinformatics
• Bioinformatics
• Network science

Background:

• Drug target similarity is traditionally assessed using sequence or structural data.
• Chemoinformatics offers an alternative by analyzing target similarity through their associated ligands.

Purpose of the Study:

• To compare chemoinformatics-derived target similarities with bioinformatics-derived ones.
• To evaluate the stability of ligand networks against varying chemoinformatics metrics.
• To determine the most reliable network for predicting drug pharmacology.

Main Methods:

• Calculated similarities between hundreds of drug targets and their ligands.
• Constructed networks based on BLAST sequence similarity (bioinformatics) and ligand-set similarity using Similarity Ensemble Approach (SEA) or Bayesian statistics (chemoinformatics).

Main Results:

• Bioinformatics and chemoinformatics networks showed substantial differences; high sequence similarity rarely correlated with high ligand-set similarity.
• Chemoinformatics networks demonstrated stability across different ligand-set similarity calculation methods and chemical representations.
• Ligand-based networks exhibited superior organization and 'small-world' and 'broad-scale' properties compared to bioinformatics networks.

Conclusions:

• Ligand-based chemoinformatics networks provide a more stable and organized framework for understanding drug target relationships.
• These chemoinformatics networks show greater potential for accurate pharmacological predictions than traditional bioinformatics approaches.