Druggable protein prediction using a multi-canal deep convolutional neural network based on autocovariance method.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Accurate identification and positioning of drug targets are crucial for pharmaceutical research and development.
• Traditional drug discovery methods can be time-consuming and costly.
• Machine learning offers a faster, more cost-effective approach to drug target identification.

Purpose of the Study:

• To develop intelligent classification systems for predicting druggable protein classes using machine learning.
• To propose and evaluate two distinct deep learning models for this prediction task.

Main Methods:

• Protein sequences were translated based on six physicochemical properties of amino acids.
• The autocovariance method was applied to convert sequences into fixed-length input vectors.
• Two deep learning architectures were employed: deep stacked sparse auto-encoders (DSSAEs) and a deep convolutional neural network (CNN).

Main Results:

• The deep convolutional neural network model demonstrated superior performance compared to previous studies in classifying druggable proteins.
• The proposed approach achieved high performance metrics: 96.92% sensitivity, 99.51% specificity, and 98.29% accuracy.

Conclusions:

• Deep learning models, particularly the CNN, are effective for intelligent classification of druggable proteins.
• This machine learning-driven approach enhances the efficiency and accuracy of drug target identification in pharmaceutical research.