SnapKin

SnapKin predicts kinase substrates and phosphorylation sites from mass spectrometry (MS)-based phosphoproteomic data using an ensemble deep learning approach to improve substrate prediction accuracy.

Key Features:

• Ensemble Deep Learning (snapshot ensemble): Employs a snapshot ensemble learning algorithm that integrates multiple deep learning models to enhance predictive performance and model stability for kinase-substrate prediction.
• Pseudo-Positive Learning Strategy: Uses a pseudo-positive learning strategy to mitigate small sample size issues arising from limited experimentally validated kinase substrates.
• Data Re-Sampling Based Ensemble Learning: Incorporates data re-sampling techniques within the ensemble to stabilize models and address high noise levels in phosphoproteomics datasets.
• Utilization of Large Phosphoproteomics Datasets: Developed and evaluated using seven large-scale phosphoproteomics datasets, including six previously published datasets and an additional muscle differentiation dataset.

Scientific Applications:

• Kinase-substrate identification: Predicts kinase-specific substrates and phosphorylation sites from complex phosphoproteomic measurements.
• Signaling pathway analysis: Supports elucidation of cellular signaling pathways and protein interaction networks by providing more accurate phosphorylation site assignments.
• Proteomics research on cell differentiation: Applied to studies of muscle differentiation using a dedicated muscle differentiation phosphoproteomics dataset.
• Disease mechanism and therapeutic target research: Aids investigation of disease mechanisms and the identification of candidate therapeutic targets through improved substrate prediction.

Methodology:

Integrates traditional and deep learning models into a snapshot ensemble framework and applies pseudo-positive learning and data re-sampling strategies for training on MS-based phosphoproteomic data.