Deep Neural Network-Assisted Drug Recommendation Systems for Identifying Potential Drug-Target Interactions

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Traditional drug-target interaction (DTI) identification is labor-intensive and low-throughput.
• In silico methods offer a more efficient alternative for discovering novel DTIs.
• Machine learning approaches are increasingly vital for analyzing complex biological data.

Purpose of the Study:

• To develop and validate a machine learning-based multiclass classification workflow for predicting DTIs.
• To segregate drug-target pairs into active, inactive, and intermediate interaction categories.
• To identify critical features governing drug-target relationships.

Main Methods:

• Utilized machine learning for a multiclass classification of drug-target pairs.
• Transformed drug molecules, protein sequences, and molecular descriptors into machine-interpretable embeddings.
• Employed tools like CHEMBL, iFeature, and a deep neural network-assisted drug recommendation (dNNDR)-featx for data processing.

Main Results:

• Trained models on large-scale DTI datasets, achieving improved performance over baseline methods.
• External validation demonstrated the predictive power of att-biLSTM and gCNN models for novel DTIs.
• The proposed models significantly outperformed competing methods on an independent dataset.

Conclusions:

• The developed machine learning workflow effectively predicts novel drug-target interactions.
• The methodology provides insights into key features driving drug-target relationships.
• Validated novel interactions align with existing experimental and computational evidence, supporting the dNNDR approach.