SecProCT: In Silico Prediction of Human Secretory Proteins Based on Capsule Network and Transformer

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Proteomics

Background:

• Secretory proteins in body fluids are key biomarkers for disease diagnosis.
• Current prediction methods rely heavily on protein features, limiting their applicability.
• Need for sequence-based prediction models that bypass feature engineering.

Purpose of the Study:

• To develop a novel deep learning model for predicting secretory proteins.
• To utilize only amino acid sequences, reducing reliance on complex feature extraction.
• To enhance the accuracy of identifying blood and saliva secretory proteins for diagnostics.

Main Methods:

• Proposed SecProCT, a deep learning model integrating capsule networks and transformer architectures.
• Input data consists solely of amino acid sequences.
• Model validation through cross-validation and independent test sets.

Main Results:

• Achieved high accuracy: 0.921 (cross-validation) and 0.917 (test set) for blood-secretory proteins.
• Achieved high accuracy: 0.892 (cross-validation) and 0.905 (test set) for saliva-secretory proteins.
• Outperformed conventional machine learning and other deep learning methods in biological sequence analysis.

Conclusions:

• SecProCT offers a superior, sequence-based approach for secretory protein prediction.
• The model effectively identifies experimentally verified secretory proteins and cancer biomarkers.
• This method advances non-invasive disease diagnostics using blood and saliva.