ResNetKhib

ResNetKhib: Deep learning-based prediction of cell type-specific lysine 2-hydroxyisobutylation sites

ResNetKhib predicts cell type-specific lysine 2-hydroxyisobutylation (Khib) sites in protein sequences using a residual network-based deep learning framework. It models Khib modifications associated with gene transcription, chromatin function regulation, purine metabolism, the pentose phosphate pathway, and glycolysis/gluconeogenesis across multiple species and cell types.

Key Features:

• Cell Type-Specific Prediction: Predicts Khib sites in four human cell types, mouse liver cells, and three rice cell types.
• Residual Network Architecture: Implements one-dimensional convolutional neural networks with ResNet-inspired residual connections.
• Transfer Learning Integration: Applies transfer learning to improve predictive performance across species and biological contexts.
• Performance Benchmarking: Evaluated using 10-fold cross-validation and independent testing; achieves AUC-ROC values ranging from 0.807 to 0.901 and outperforms random forest predictors.

Scientific Applications:

• Post-Translational Modification Analysis: Identifies Khib modification sites to support mechanistic studies of protein regulation and metabolic pathway involvement.

Methodology:

ResNetKhib employs one-dimensional convolutional neural networks with residual connections inspired by ResNet architecture and incorporates transfer learning. Model performance is assessed using 10-fold cross-validation and independent test datasets across human, mouse, and rice cell types.