GASIDN

GASIDN identifies sub-Golgi proteins through multi-scale feature fusion of protein sequence and spatial structure information.

Key Features:

• Multi-scale feature fusion: Combines protein sequence features with contact map information derived from AlphaFold2.
• 1D convolution module: Extracts multi-dimension features from protein sequences.
• Graph learning module: Processes contact maps constructed from AlphaFold2 structural predictions.
• SeqVec initialization: Utilizes the deep representation learning model SeqVec to initialize protein sequences.
• Cross-organelle generalizability: Applicable to identification of proteins from other organelles including plant vacuoles and peroxisomes.

Scientific Applications:

• Sub-Golgi protein identification: Classifies and localizes protein types within the Golgi apparatus to support research on neurodegenerative diseases linked to Golgi dysfunction.
• Plant vacuole protein identification: Extends to identification of proteins in plant vacuoles.
• Peroxisome protein identification: Extends to identification of proteins in peroxisomes.

Methodology:

GASIDN extracts multi-dimension features using a 1D convolution module on protein sequences and a graph learning module on contact maps constructed from AlphaFold2, with protein sequences initialized by the SeqVec deep representation learning model.