DTL-DephosSite: Deep Transfer Learning Based Approach to Predict Dephosphorylation Sites

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• Protein phosphorylation is a key post-translational modification regulating cellular processes.
• Existing computational tools primarily focus on predicting phosphorylation sites, neglecting dephosphorylation.
• Dephosphorylation is crucial for regulating protein phosphorylation status and cellular functions.

Purpose of the Study:

• To develop a computational model for predicting dephosphorylation sites.
• To address the knowledge gap in dephosphorylation site prediction tools.
• To enhance the understanding of dephosphorylation's role in cellular regulation.

Main Methods:

• Employed a transfer learning strategy.
• Developed a deep learning-based model named DTL-DephosSite.
• Utilized independent test results for model evaluation.

Main Results:

• DTL-DephosSite achieved high efficiency scores for phosphoserine/phosphothreonine residues (SN: 84%, SP: 84%, MCC: 0.68).
• The model also demonstrated strong performance for phosphotyrosine residues (SN: 75%, SP: 88%, MCC: 0.64).
• The model effectively predicts sites likely to be dephosphorylated.

Conclusions:

• DTL-DephosSite successfully predicts dephosphorylation sites using a deep learning approach.
• The developed model fills a critical gap in computational tools for studying protein modification.
• This advancement aids in understanding the dynamic regulation of cellular processes through dephosphorylation.