GPS-Uber

GPS-Uber predicts general and E3-specific lysine ubiquitination sites by integrating sequence and structural features with machine learning to identify modification sites relevant to protein regulation and disease.

Key Features:

• Comprehensive Dataset: A manually curated dataset of 1311 experimentally validated site-specific E3–substrate relations organized into clusters based on corresponding E3s at different levels.
• Integrated Feature Set: Prediction of general ubiquitination sites uses a diverse feature set including 10 types of sequence features and structure features.
• Advanced Algorithms: Employs penalized logistic regression, deep neural network (DNN), and convolutional neural network (CNN) approaches for model training.
• Competitive Accuracy: The general model achieves an area under the curve (AUC) of 0.7649.
• Transfer Learning for E3-Specific Predictions: Uses transfer learning to create 112 distinct E3-specific predictors tailored to individual E3 clusters.

Scientific Applications:

• Cancer-associated Ubiquitination Analysis: Systematic analysis of human cancer-associated ubiquitination events to identify potential ubiquitination sites linked to cancer.

Methodology:

Hybrid-learning framework combining sequence and structural data with penalized logistic regression, deep neural networks, convolutional neural networks, and transfer learning for general and E3-specific model training.