PhosAF: An integrated deep learning architecture for predicting protein phosphorylation sites with AlphaFold2...

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Phosphorylation is a key post-translational modification regulating numerous cellular functions.
• Experimental identification of phosphorylation sites is labor-intensive and time-consuming.
• Current deep learning predictors often overlook crucial protein structural information.

Purpose of the Study:

• To develop an advanced deep learning model for accurate human phosphorylation site prediction.
• To leverage both protein sequence and structural information for enhanced prediction accuracy.
• To improve upon existing methods by incorporating structural data from AlphaFold2 predictions.

Main Methods:

• Developed PhosAF, an integrated deep learning architecture combining CMA-Net and MFC-Net.
• CMA-Net utilizes convolutional neural networks and multi-head attention for sequence feature analysis.
• MFC-Net employs deep neural networks to process evolutionary and structural features, with novel negative sample generation using protein secondary structures.

Main Results:

• PhosAF effectively integrates sequence and structure information for phosphorylation site prediction.
• The model demonstrates superior performance compared to existing state-of-the-art methods.
• Validation through independent test data and case studies confirms PhosAF's effectiveness.

Conclusions:

• PhosAF represents a significant advancement in predicting human phosphorylation sites.
• Integrating structural information alongside sequence data improves prediction accuracy.
• The developed model offers a more efficient and accurate alternative to experimental methods.