BCM-DTI: A fragment-oriented method for drug-target interaction prediction using deep learning

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) identification is crucial but costly due to vast molecular and protein spaces.
• Current deep learning methods often use whole molecule/protein features, neglecting the importance of specific substructures in pharmacological reactions.
• Existing substructure-focused approaches may only consider limited fragment types, such as functional groups.

Purpose of the Study:

• To develop an advanced deep learning framework for predicting drug-target interactions (DTI).
• To address the limitations of existing DTI prediction methods by incorporating diverse molecular and protein substructures.
• To enhance the accuracy and efficiency of DTI prediction in drug discovery.

Main Methods:

• Proposed BCM-DTI, an end-to-end framework for DTI prediction.
• Integrated diverse fragment types, including branch chains, common substructures, and motifs.
• Employed a Convolutional Neural Network (CNN) based feature learning module to capture synergistic effects between fragments.

Main Results:

• BCM-DTI was implemented and evaluated on four public datasets.
• The framework demonstrated superior performance compared to state-of-the-art DTI prediction approaches.
• BCM-DTI achieved higher accuracy with a reduced training cost.

Conclusions:

• BCM-DTI effectively predicts drug-target interactions by considering multiple types of molecular and protein fragments.
• The proposed framework offers a more theoretically grounded and computationally efficient approach to DTI prediction.
• BCM-DTI represents a significant advancement in computational drug discovery and development.