PhosBERT: A self-supervised learning model for identifying phosphorylation sites in SARS-CoV-2-infected human cells

Area of Science:

• Computational Biology
• Virology
• Biochemistry

Background:

• Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes COVID-19, leading to protein post-translational modification dysregulation in host cells.
• Understanding these modifications, particularly phosphorylation sites, is crucial for elucidating SARS-CoV-2 pathogenesis and identifying antiviral therapies.

Purpose of the Study:

• To develop a cost-effective and high-precision computational strategy for identifying SARS-CoV-2-infected phosphorylation sites.
• To enhance the understanding of viral-host interactions at the molecular level.

Main Methods:

• Implementation of a custom neural network model, PhosBERT, based on the pre-trained protein language model ProtBert.
• Training and validation using serine (S), threonine (T), and tyrosine (Y) phosphorylation datasets with 5-fold cross-validation.

Main Results:

• PhosBERT achieved high accuracy in identifying S/T phosphorylation sites (81.9% accuracy, 0.896 AUC).
• PhosBERT demonstrated high prediction accuracy for Y phosphorylation sites (87.1% accuracy, 0.902 AUC).
• The model exhibited good prediction ability and stability in independent validation.

Conclusions:

• PhosBERT provides a novel and effective computational approach for studying SARS-CoV-2-associated phosphorylation sites.
• This tool can significantly contribute to understanding COVID-19 mechanisms and accelerate antiviral drug screening.