SPVec: A Word2vec-Inspired Feature Representation Method for Drug-Target Interaction Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Accurate identification of drug-target interactions (DTIs) is crucial for drug discovery.
• Traditional experimental methods for DTI identification are costly and time-consuming.
• Existing machine learning (ML) approaches are limited by the need for extensive human expertise in feature extraction.

Purpose of the Study:

• To develop a novel, unsupervised method (SPVec) for automatic feature extraction from raw molecular and protein data.
• To improve the efficiency and accuracy of machine learning-based DTI prediction.
• To reduce reliance on manual feature engineering in drug discovery.

Main Methods:

• Proposed SPVec, an unsupervised representation learning method inspired by Word2vec.
• Applied SPVec to convert SMILES strings (compounds) and protein sequences into continuous, lower-dimensional vectors.
• Evaluated SPVec using state-of-the-art ML classifiers (Gradient Boosting Decision Tree, Random Forest, Deep Neural Network) on BindingDB and DrugBank datasets.

Main Results:

• SPVec generated informative vectors, with similar compounds/proteins clustering together in visualization.
• SPVec outperformed manually designed features (MACCS, AAC) in DTI prediction accuracy.
• Validated SPVec's performance and robustness on independent test sets from DrugBank.
• Successfully predicted potential novel DTIs, with external evidence supporting two of the top five predictions.

Conclusions:

• SPVec offers an effective and efficient automated approach for DTI identification.
• The method reduces the need for manual feature extraction, accelerating ML applications in drug discovery.
• SPVec shows promise for drug reprofiling and the discovery of new therapeutic applications.