DeepKinZero: zero-shot learning for predicting kinase-phosphosite associations involving understudied kinases

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Protein phosphorylation is crucial for cell signaling, with kinases catalyzing this process.
• Dysregulated phosphorylation is linked to diseases like cancer.
• Most human phosphosites lack identified kinases, hindering research.

Purpose of the Study:

• To develop a computational method for predicting kinases acting on phosphosites, especially for kinases with limited or no known targets.
• To address the challenge of identifying kinase-phosphosite relationships for understudied kinases.

Main Methods:

• Introduced DeepKinZero, a zero-shot learning model.
• Utilized a bidirectional recurrent neural network to learn kinase-specific amino acid preferences.
• Transferred knowledge from well-characterized kinases to those with unknown targets.

Main Results:

• DeepKinZero demonstrated significant accuracy improvements for kinases with no known phosphosites.
• Outperformed baseline models and existing methods in predicting kinase-target relationships.
• Successfully identified kinases for previously uncharacterized phosphosites.

Conclusions:

• DeepKinZero is the first zero-shot learning approach for kinase-phosphosite prediction.
• The method enhances the understanding of understudied kinases and contributes to mapping the phosphoproteome.
• Enables broader phosphoproteome analysis by providing kinase information for uncharacterized sites.