Predicting drug-target interactions using restricted Boltzmann machines.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• In silico drug-target interaction (DTI) prediction is crucial for drug discovery and repurposing.
• Existing network-based DTI prediction methods often overlook diverse interaction types, limiting their predictive power.
• Integrating varied DTI information can enhance prediction accuracy and elucidate drug mechanisms.

Purpose of the Study:

• To develop a novel machine learning approach for predicting multiple types of drug-target interactions (DTIs).
• To improve the prediction of unknown drug-target relationships and drug modes of action.
• To leverage diverse DTI data for advancing drug discovery and repositioning.

Main Methods:

• A machine learning approach utilizing a two-layer restricted Boltzmann machine (RBM) was developed.
• The RBM model was trained using a practical learning algorithm to predict DTIs.
• The approach integrates multiple types of DTIs, unlike previous methods that often use single interaction types or mixed data.

Main Results:

• The proposed RBM model achieved excellent performance in predicting various DTIs, with an area under the precision-recall curve reaching 89.6%.
• Integrating multiple DTI types significantly outperformed methods using single interaction types or undifferentiated mixed data.
• The approach successfully inferred a high proportion of novel DTIs, many validated by existing experimental data.

Conclusions:

• The developed machine learning approach effectively predicts diverse drug-target interactions (DTIs).
• Integrating multiple DTI types offers a significant advantage over traditional methods for DTI prediction.
• This approach holds practical relevance for drug repositioning and accelerating the drug discovery pipeline.