Deep Neural Network Framework Based on Word Embedding for Protein Glutarylation Sites Prediction.

Area of Science:

• Biochemistry
• Computational Biology
• Proteomics

Background:

• Dysregulation of glutarylation is linked to human diseases including diabetes, cancer, and glutaric aciduria type I.
• Accurate identification and characterization of glutarylation sites are vital for modification-specific proteomics.

Purpose of the Study:

• To propose a novel deep neural network (DNN) framework utilizing word embedding techniques for predicting glutarylation sites.
• To evaluate and compare the performance of various DNN models and word embedding methods for protein sequence data representation.

Main Methods:

• Implementation of multiple deep neural network models.
• Extensive experimental comparison of different word embedding techniques for protein sequence representation.
• Performance evaluation using metrics such as accuracy, specificity, sensitivity, and correlation coefficient.

Main Results:

• The proposed DNN framework significantly improves protein sequence representation and glutarylation site prediction.
• Achieved higher accuracy and confidence rates compared to previous methods.
• Embedding techniques proved more effective than pre-trained word embeddings for glutarylation sequence representation.
• Outperformed traditional methods with accuracy (0.79), specificity (0.89), sensitivity (0.59), and correlation coefficient (0.51).

Conclusions:

• The novel DNN framework offers a powerful tool for glutarylation site prediction.
• Demonstrates potential for identifying novel glutarylation sites and elucidating connections between glutarylation and lysine modifications.
• Highlights the efficacy of word embedding techniques in enhancing protein sequence analysis for post-translational modification studies.