A deep learning-based method for drug-target interaction prediction based on long short-term memory neural network.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Traditional experimental methods for identifying drug-target interactions (DTIs) are limited by throughput, precision, and cost.
• There is a critical need for efficient computational approaches to predict potential DTIs.
• Accurate DTI prediction is essential for modern drug discovery and development.

Purpose of the Study:

• To develop and validate a deep learning-based computational model for predicting Drug-Target Interactions (DTIs).
• To enhance the accuracy and efficiency of DTI prediction compared to existing methods.
• To demonstrate the model's effectiveness, even with limited data.

Main Methods:

• Extracted protein evolutionary features using Position Specific Scoring Matrix (PSSM) and Legendre Moment (LM).
• Combined protein features with drug molecular substructure fingerprints to create feature vectors for drug-target pairs.
• Employed Sparse Principal Component Analysis (SPCA) for feature compression into a uniform vector space.
• Utilized a Deep Long Short-Term Memory (DeepLSTM) network for DTI prediction.

Main Results:

• Achieved high prediction performance with Area Under the Curve (AUC) values of 0.9951, 0.9705, 0.9951, and 0.9206 on four key drug-target datasets.
• Demonstrated the proposed characterization scheme's superiority in feature expression and recognition.
• Confirmed the model's robust performance even on small datasets.

Conclusions:

• The developed deep learning approach significantly outperforms state-of-the-art drug-target predictors.
• This study is the first to explore the potential of deep learning with memory and Turing completeness for DTI prediction.
• The findings highlight a promising computational strategy for accelerating drug discovery by accurately predicting DTIs.