CapsNet-SSP

CapsNet-SSP predicts human saliva-secretory proteins from protein sequence data using a multilane capsule network to identify candidate salivary biomarkers.

Key Features:

• End-to-End Deep Learning Model: CapsNet-SSP employs a multilane capsule network (CapsNet) architecture with differently sized convolution kernels to learn feature representations directly from sequence data without requiring annotated protein features.
• Automatic Feature Representation: The model automatically extracts and represents sequence-derived features, bypassing the need for hand-crafted or pre-labeled protein features.
• Superior Performance: CapsNet-SSP outperforms existing methods based on traditional machine learning and demonstrates superior performance relative to other deep learning architectures used in biological sequence analysis.
• Validation Against Known Biomarkers: Predictions have been validated against known human saliva-secretory proteins and established salivary protein biomarkers associated with cancer, showing statistical significance.

Scientific Applications:

• Salivary biomarker discovery: Enables identification of candidate saliva-secretory proteins as potential diagnostic or prognostic biomarkers from sequence data, leveraging saliva's noninvasive sampling.
• Cancer-related biomarker analysis: Applied to identify and evaluate salivary protein biomarkers associated with cancer.
• Transcriptome and proteomics follow-up: Provides candidate proteins for follow-up validation in transcriptome or proteomics analyses.
• Study of salivary gland-related diseases: Supports exploration of proteins secreted from tissues proximal and distal to salivary glands in disease contexts.

Methodology:

CapsNet-SSP uses a multilane capsule network processing biological sequence data with various convolution kernel sizes to capture sequence patterns and learn feature representations without reliance on pre-labeled features.