CGBVS-DNN: Prediction of Compound-protein Interactions Based on Deep Learning

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Accurate prediction of compound-protein interactions (CPIs) is crucial for efficient in-silico drug design and screening.
• Previous methods like chemical genomics-based virtual screening (CGBVS) using support vector machines (SVMs) face scalability challenges with large datasets due to high computational demands.

Purpose of the Study:

• To develop a more scalable and efficient computational method for predicting CPIs.
• To address the limitations of existing SVM-based approaches when handling massive CPI datasets.

Main Methods:

• Proposed CGBVS-DNN, utilizing deep neural networks (DNNs) instead of SVMs for CPI prediction.
• Implemented a mini-batch training approach for DNNs, enabling efficient learning from large datasets without requiring all data at once.

Main Results:

• CGBVS-DNN demonstrated superior performance compared to the original CGBVS method using a quarter million CPIs.
• Cross-validation results indicated a high accuracy of up to 98.2% (σ<0.01) for CGBVS-DNN with 4.5 million CPIs.

Conclusions:

• Deep learning, specifically CGBVS-DNN, offers a powerful and scalable solution for computational prediction of CPIs.
• The CGBVS-DNN method significantly enhances the efficiency and accuracy of virtual screening in drug discovery pipelines.