Prediction of Peptide Detectability Based on CapsNet and Convolutional Block Attention Module

Area of Science:

• Proteomics
• Computational Biology
• Bioinformatics

Background:

• Mass spectrometry experiments in proteomics face challenges with reproducibility and random peptide identification due to complex sampling.
• Accurate prediction of peptide detectability is crucial for optimizing experimental results and enhancing data reliability.

Purpose of the Study:

• To develop a novel method for predicting peptide detectability using advanced deep learning techniques.
• To improve the accuracy and reproducibility of peptide identification and quantification in mass spectrometry.

Main Methods:

• A capsule network (CapsNet) integrated with a convolutional block attention module (CBAM) was employed.
• Peptide features including residue conical coordinate (RCC), amino acid composition (AAC), dipeptide composition (DPC), and sequence embedding code (SEC) were extracted and categorized into biological and sequence features.
• These features were input into the CapsNet, with CBAM enhancing feature learning by assigning channel and space weights.

Main Results:

• The proposed CapsNet-CBAM method demonstrated superior performance compared to existing popular methods in most assessment metrics.
• The integration of CBAM effectively enhanced feature learning and network training, leading to improved prediction accuracy.

Conclusions:

• The novel CapsNet-CBAM method offers a significant advancement in predicting peptide detectability.
• This approach holds promise for enhancing the accuracy, reproducibility, and overall reliability of proteomics data analysis.