AutoDTI++: deep unsupervised learning for DTI prediction by autoencoders.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug-target interactions (DTIs) are crucial for drug discovery but experimentally identifying them is costly and time-consuming.
• Computational methods, especially deep learning, are increasingly used to predict DTIs and reduce experimental efforts.
• Traditional methods often rely on protein sequences and drug molecular structures, alongside supervised learning.

Purpose of the Study:

• To propose a novel deep unsupervised learning method for predicting drug-target interactions.
• To enhance the accuracy and efficiency of computational drug-target interaction prediction.
• To address the sparsity challenge in drug-target interaction matrices.

Main Methods:

• The proposed AutoDTI++ method employs deep unsupervised learning.
• It involves pre-processing the sparse drug-target interaction matrix using drug fingerprints.
• The core AutoDTI approach is applied, followed by post-processing of the model's output.

Main Results:

• The AutoDTI++ method demonstrated improved prediction performance compared to existing algorithms.
• Experiments were conducted using standard datasets: Nuclear Receptors, GPCRs, Ion channels, and Enzymes.
• The method achieved high accuracy, validated through five repetitions of tenfold cross-validation.

Conclusions:

• AutoDTI++ offers a robust computational approach for predicting drug-target interactions.
• The deep unsupervised learning strategy effectively handles sparse interaction data.
• The method shows significant potential for accelerating drug discovery by accurately identifying potential drug-target pairs.