Identification of phosphorylation site using S-padding strategy based convolutional neural network.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Abnormal protein phosphorylation is linked to various human diseases.
• Accurate identification of phosphorylation sites is crucial for disease research.
• Current deep learning models for phosphorylation site prediction can be complex and limited in application.

Purpose of the Study:

• To develop an enhanced deep learning method for accurate phosphorylation site prediction.
• To improve the usability and applicability of phosphorylation site prediction models.
• To leverage deep learning for a better understanding of phosphorylation mechanisms.

Main Methods:

• An enhanced deep learning method utilizing a convolutional neural network (CNN) with an S-padding strategy.
• The S-padding strategy creates a 3D matrix from amino acid sequences, capturing extension information.
• A 2D-CNN model abstracts comprehensive features from protein sequences for site prediction.

Main Results:

• Achieved 89.68% accuracy on serine/threonine sites and 88.16% on tyrosine sites in human datasets.
• Demonstrated accuracy, sensitivity, and specificity over 0.85 for phosphorylation site prediction across different organisms.
• Outperformed existing models in accuracy and AUC, with potential for further improvement using more training data.

Conclusions:

• The proposed method enables proteome-wide phosphorylation site predictions.
• This approach enhances the potential of protein phosphorylation site identification.
• The findings offer insights into phosphorylation mechanisms and associated diseases.