Adaptive learning embedding features to improve the predictive performance of SARS-CoV-2 phosphorylation sites

Area of Science:

• Virology
• Molecular Biology
• Bioinformatics

Background:

• Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission caused a global health crisis.
• Understanding host cell pathway alterations during SARS-CoV-2 infection is crucial.
• Existing phosphorylation site prediction tools for SARS-CoV-2 lack accuracy and efficiency.

Purpose of the Study:

• To analyze biological functions affected by SARS-CoV-2 infection in human lung cells.
• To develop a novel deep learning predictor for identifying phosphorylation sites in SARS-CoV-2 infected host cells.
• To improve the accuracy and efficiency of phosphorylation site prediction.

Main Methods:

• Comprehensive biological function analysis of SARS-CoV-2 infected A549 lung epithelial cells.
• Development of PSPred-ALE, a deep learning predictor utilizing self-adaptive learning embedding and multihead attention.
• Extraction of context sequential features from protein sequences for improved prediction.

Main Results:

• Dramatic changes in host cell protein phosphorylation pathways were observed.
• PSPred-ALE demonstrated superior performance compared to existing state-of-the-art prediction tools.
• Self-adaptive learning embedding features outperformed hand-crafted statistical features.

Conclusions:

• PSPred-ALE effectively identifies phosphorylation sites in SARS-CoV-2 infected host cells.
• The tool assists biomedical scientists in elucidating phosphorylation mechanisms during SARS-CoV-2 infection.
• PSPred-ALE offers a robust and accurate solution for phosphorylation site prediction.