Matrix factorization with denoising autoencoders for prediction of drug-target interactions

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-target interactions (DTI) are fundamental to new drug discovery.
• Computational methods, especially machine learning, are increasingly used for DTI prediction.
• Matrix factorization methods for DTI prediction suffer from data sparsity and cold-start issues.

Purpose of the Study:

• To propose a hybrid DTI prediction model integrating matrix factorization and denoising autoencoders.
• To leverage drug and target side information to overcome the sparsity of the interaction matrix.
• To enhance the accuracy and robustness of DTI prediction.

Main Methods:

• A two-step approach involving interaction matrix pre-processing and a hybrid model.
• Utilizing similarity matrices of drugs and targets to address interaction matrix sparsity.
• Combining matrix factorization with denoising autoencoders to learn hidden factors from side information.

Main Results:

• The proposed hybrid model demonstrated superior performance compared to existing algorithms.
• High accuracy was achieved on benchmark datasets, including Nuclear Receptors, GPCRs, Ion Channels, and Enzymes.
• Evaluation metrics included area under the receiver operating characteristic curve (AUC-ROC) and area under the precision-recall curve (AUC-PR).

Conclusions:

• The hybrid DTI prediction model effectively handles data sparsity and cold-start problems.
• The integration of side information significantly improves the accuracy of drug-target interaction prediction.
• This approach offers a promising computational strategy for accelerating drug discovery.