PhosIDN

PhosIDN predicts protein phosphorylation sites by integrating protein sequence and protein–protein interaction (PPI) data within a deep neural network to improve phosphorylation-site identification.

Key Features:

• Integrated Deep Learning Architecture: PhosIDN employs a deep neural network architecture that integrates protein sequence and protein–protein interaction (PPI) data to enhance phosphorylation-site prediction accuracy.
• Sequence Feature Encoding Sub-Network: A dedicated sub-network encodes sequence features to capture local patterns and long-range dependencies within protein sequences, addressing limitations of convolutional neural networks (CNNs).
• PPI Feature Extraction: A multi-layer deep neural sub-network extracts features from PPI data to represent interaction context relevant to phosphorylation.
• Heterogeneous Feature Combination Sub-Network: A sub-network combines sequence-derived and PPI-derived features to model complex associations between heterogeneous data types for prediction.

Scientific Applications:

• Protein function and regulation: Prediction of phosphorylation sites to support studies of protein function and regulatory mechanisms.
• Cellular signaling pathways: Mapping phosphorylation events within cellular signaling pathways.
• Disease mechanism analysis (such as cancer): Investigating altered phosphorylation patterns in disease mechanisms, including cancer.
• Kinase-targeted drug discovery: Informing drug discovery efforts targeting kinase activities by identifying candidate phosphorylation sites.

Methodology:

PhosIDN uses a deep neural network comprising a sequence feature encoding sub-network that captures local and long-range sequence dependencies, a multi-layer deep neural PPI feature extraction sub-network, and a heterogeneous feature combination sub-network to integrate sequence and PPI features.