A Novel Approach for Drug-Target Interactions Prediction Based on Multimodal Deep Autoencoder

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Drug-target interactions (DTIs) are crucial for therapeutic effects and disease treatment.
• Accurate DTI prediction requires integrating multiple drug and target similarity measures.
• Existing methods often overlook global structural information and non-linear feature relationships.

Purpose of the Study:

• To propose a novel computational approach, MDADTI, for accurate drug-target interaction prediction.
• To address limitations of existing methods by incorporating global structure information and automatic feature learning.
• To enhance the prediction of drug-target interactions for improved disease therapy.

Main Methods:

• MDADTI utilizes random walk with restart and positive pointwise mutual information to compute topological similarity matrices.
• A multimodal deep autoencoder fuses these matrices, automatically learning low-dimensional drug and target features.
• A deep neural network is employed for the final DTI prediction.

Main Results:

• MDADTI demonstrated superior performance compared to four baseline methods across three cross-validation settings (5x10-fold CV).
• The method effectively identified unknown drug-target interactions, validated against six reference databases.
• The approach successfully captured global structure information and learned non-linear feature relationships.

Conclusions:

• MDADTI offers a significant advancement in predicting drug-target interactions.
• The method's ability to integrate global structure and learn features enhances DTI prediction accuracy.
• MDADTI holds promise for accelerating drug discovery and development by identifying novel therapeutic targets.