Shaping the interaction landscape of bioactive molecules.

Area of Science:

• Computational chemistry
• Cheminformatics
• Pharmacology

Background:

• Many bioactive molecules interact with proteins, but their specific targets are often unknown.
• Most bioactive molecules exhibit polypharmacology, targeting multiple proteins, with many targets poorly characterized.
• Computational target prediction using ligand similarity is vital for identifying potential targets and explaining molecule side effects.

Purpose of the Study:

• To develop and validate a computational method for accurately predicting bioactive molecule targets.
• To enhance the understanding of ligand-target interactions in drug discovery.

Main Methods:

• Utilized a reference dataset of 224,412 molecules with known activity against 1700 human proteins.
• Employed a combined approach using various chemical similarity measures, including structural and shape-based metrics.
• Analyzed the optimal combination of similarity measures based on molecular properties.

Main Results:

• Accurate bioactive molecule target prediction was achieved by integrating diverse chemical similarity measures.
• The combined similarity approach proved particularly effective for novel molecules lacking known ligands within the same chemical series or scaffold.
• Optimal similarity measure combinations varied depending on molecular properties, such as heavy atom count.

Conclusions:

• Combining structural and shape-based chemical similarity is a powerful strategy for accurate bioactive molecule target prediction.
• This integrated approach is especially valuable for identifying targets of novel chemical entities.
• The findings underscore the importance of employing multiple similarity metrics for robust ligand-target prediction in drug discovery.