DeepCSO: A Deep-Learning Network Approach to Predicting Cysteine S-Sulphenylation Sites.

Area of Science:

• Biochemistry
• Proteomics
• Bioinformatics

Background:

• Cysteine S-sulphenylation (CSO) is a significant post-translational modification (PTM) regulating protein function and signaling networks.
• Existing CSO prediction tools are limited, primarily focusing on *Homo sapiens* and lacking cross-species applicability.
• Recent proteomic studies have expanded the identified CSO sites, necessitating updated and broader prediction models.

Purpose of the Study:

• To develop advanced prediction models for cysteine S-sulphenylation (CSO) sites applicable across various species.
• To investigate the characteristics of CSO modifications in a multi-species context.
• To create a user-friendly online service for CSO site prediction.

Main Methods:

• Development and comparison of various classification models, including traditional machine learning and deep learning approaches.
• Utilizing long short-term memory (LSTM) networks with word-embedding encoding for enhanced prediction.
• Training and validating models on an enlarged, multi-species dataset of identified CSO sites.

Main Results:

• The LSTM-based model demonstrated superior performance compared to traditional and other deep-learning models across different species.
• The area under the receiver operating characteristic (ROC) curve for the LSTM model ranged from 0.82 to 0.85, outperforming existing CSO predictors.
• A general prediction model integrating data from multiple species exhibited high universality and effectiveness.

Conclusions:

• The developed LSTM model, DeepCSO, provides a robust and accurate method for predicting CSO sites across diverse species.
• DeepCSO offers improved prediction accuracy and broader applicability than previously available tools.
• The online prediction service DeepCSO facilitates research on cysteine S-sulphenylation by providing species-specific and general prediction models.