CapsNet-SSP: multilane capsule network for predicting human saliva-secretory proteins

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Saliva biomarkers offer noninvasive advantages over blood or urine for clinical diagnostics.
• Identifying human saliva-secretory proteins is crucial for developing new clinical tests.
• Existing methods for predicting saliva-secretory proteins often rely on feature engineering and conventional machine learning.

Purpose of the Study:

• To develop an advanced deep learning model for predicting saliva-secretory proteins using only sequence information.
• To improve the accuracy and efficiency of identifying potential salivary protein biomarkers.
• To provide a computational tool for researchers investigating salivary gland-related diseases.

Main Methods:

• An end-to-end deep learning model, multilane capsule network (CapsNet), was developed for saliva-secretory protein identification.
• The model, CapsNet-SSP, utilizes differently sized convolution kernels to process protein sequence data.
• Performance was evaluated against conventional machine learning algorithms and other deep learning architectures.

Main Results:

• The CapsNet-SSP model demonstrated superior performance compared to existing conventional machine learning methods.
• It also outperformed other state-of-the-art deep learning models commonly used for biological sequence analysis.
• Predicted saliva-secretory proteins showed statistical significance when compared with known salivary cancer biomarkers.

Conclusions:

• A novel CapsNet-based model (CapsNet-SSP) effectively identifies saliva-secretory proteins from sequence data.
• The model offers improved performance over existing prediction methods.
• A web server for CapsNet-SSP is available, aiding researchers in identifying salivary protein biomarkers for disease research.