DeepCleave

DeepCleave predicts protease-specific substrates and their cleavage sites from protein substrate sequences using convolutional neural networks and transfer learning to characterize cleavage patterns for caspases and matrix metalloproteases.

Key Features:

• High-Quality Cleavage Site Features: Cleavage-site-specific features are extracted from substrate sequences via deep learning to represent sequence signals around cleavage positions.
• Convolutional Neural Networks (CNNs): CNNs model local sequence patterns and contextual information around putative cleavage sites.
• Transfer Learning: Transfer learning is applied to leverage pre-existing knowledge and improve model generalization across datasets.
• Multiple Kernels and Attention Layer: Multiple convolutional kernels and an attention layer are integrated into the network architecture for nuanced sequence analysis.
• Performance Benchmarking: Empirical evaluations report superior predictive performance relative to several state-of-the-art methods for caspase and matrix metalloprotease cleavage-site prediction.

Scientific Applications:

• Protease Substrate Identification: Identification of potential substrates and precise cleavage sites for caspases and matrix metalloproteases.
• Proteolytic Pathway Analysis: Analysis of proteolytic pathways and protease-specific cleavage patterns in protein life-cycle regulation.
• Disease-related Proteolysis Studies: Investigation of proteolytic impacts and physiological consequences of caspase and matrix metalloprotease activity in health and disease.

Methodology:

DeepCleave uses protein substrate sequence data as input, employs convolutional neural networks with multiple kernels and an attention layer, applies transfer learning, and extracts cleavage-site features through deep learning-based feature representation.