PTMFusionNet: A Deep Learning Approach for Predicting Disease Related Post-translational Modification and Classifying...

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Genomics and Proteomics

Background:

• Advanced mass spectrometry enables simultaneous detection of protein intensity and post-translational modification (PTM) data.
• Current PTM data integration with protein expression data is insufficient for comprehensive clinical applications.
• Accurate disease subtyping requires robust integration of multi-omics data.

Purpose of the Study:

• To develop a deep learning framework, PTMFusionNet, for predicting disease-related PTMs.
• To integrate predicted PTMs with protein expression data for enhanced disease subtyping.
• To improve the accuracy and robustness of disease classification using integrated proteomic and PTM data.

Main Methods:

• PTMFusionNet employs two Graph Convolutional Network (GCN) models: Layer-Attention GCN (LAGCN) for PTM potentiality scoring and Feature Weighting GCN (FWGCN) for data integration.
• LAGCN predicts the likelihood of specific PTMs occurring in disease contexts.
• FWGCN fuses PTM scores with protein expression data for classification tasks.

Main Results:

• PTMFusionNet demonstrated superior performance across three independent datasets (KIPAN, COADREAD, THCA).
• The model achieved higher accuracy, F1 score, and AUC compared to existing benchmark algorithms.
• Key PTM biomarkers relevant to disease subtyping were effectively identified.

Conclusions:

• PTMFusionNet offers a robust method for integrating PTM and protein expression data.
• The framework significantly advances disease subtyping by leveraging deep learning for multi-omics data analysis.
• This approach holds promise for improved clinical diagnosis and personalized treatment strategies.