DeepTL-Ubi

DeepTL-Ubi predicts ubiquitination sites across multiple species using deep transfer learning to transfer patterns learned from extensive human ubiquitination datasets and improve species-specific prediction performance.

Key Features:

• Deep transfer learning: Transfers learned representations from large human ubiquitination datasets to target species to mitigate limited training data.
• Cross-species prediction: Predicts ubiquitination sites across multiple species by leveraging shared sequence patterns.
• Deep learning-based feature learning: Learns predictive features directly from input data, reducing reliance on manual feature engineering.
• Training on integrated multi-source datasets: Trained and tested on ubiquitination sites integrated from multiple sources and species.
• Improved performance for small sample sizes: Enhances predictive accuracy for non-human species with limited experimental data.
• Computational alternative to mass spectrometry: Serves as a computational approach to complement mass spectrometry-based identification of ubiquitination sites.
• Reported comparative performance: Demonstrates superior performance compared to existing tools across many species.

Scientific Applications:

• Cross-species ubiquitination mapping: Enables prediction of ubiquitination sites in diverse organisms when experimental data are scarce.
• Post-translational modification studies: Supports investigations of protein regulation involving ubiquitination as a post-translational modification.
• Comparative and evolutionary analysis: Facilitates cross-species research into conserved and divergent ubiquitination patterns.

Methodology:

Uses a deep learning framework with transfer learning that trains on extensive human ubiquitination datasets, transfers learned representations to other species, and is trained and tested on integrated multi-species ubiquitination site datasets while replacing manual feature engineering with deep feature learning.