Prediction of biological targets for compounds using multiple-category Bayesian models trained on chemogenomics...

Area of Science:

• Computational chemistry
• Cheminformatics
• Pharmacology

Background:

• Accurate target identification is crucial after discovering bioactive small molecules.
• Existing experimental target fishing methods can be time-consuming.
• In silico approaches offer a rapid alternative for predicting drug targets.

Purpose of the Study:

• To develop an in silico tool for predicting potential protein targets of small molecules using only their chemical structures.
• To create a computational correlate of experimental target fishing technologies.
• To aid in the rapid identification of targets for novel compounds and deconvolve generic activities.

Main Methods:

• A multiple-category Laplacian-modified naïve Bayesian model was developed.
• The model was trained on extended-connectivity fingerprints of compounds from 964 target classes in the World Of Molecular BioAcTivity (WOMBAT) database.
• The model predicted the top three most likely protein targets for compounds in the MDL Drug Database Report (MDDR) database.

Main Results:

• The model correctly identified the target 77% of the time for compounds from 10 MDDR activity classes with known targets.
• The tool successfully deconvoluted generic therapeutic activities (e.g., 'antineoplastic') into specific protein targets.
• Demonstrated utility in predicting new targets for orphan compounds and enhancing chemogenomics databases.

Conclusions:

• The developed in silico model provides an effective means for rapid target identification based on chemical structure alone.
• This computational approach aids in drug discovery by predicting targets and refining knowledge within chemogenomics databases.
• The tool is valuable for deconvoluting broad activity annotations into specific molecular targets, advancing precision medicine.