DeepKhib: A Deep-Learning Framework for Lysine 2-Hydroxyisobutyrylation Sites Prediction.

Area of Science:

• Biochemistry and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Lysine 2-Hydroxyisobutyrylation (Khib) is a crucial post-translational modification involved in gene transcription and signal transduction.
• Understanding Khib regulatory mechanisms requires accurate identification of Khib sites.
• Existing prediction algorithms are limited in species coverage and generalizability.

Purpose of the Study:

• To develop a robust and generalizable deep-learning model for predicting lysine 2-Hydroxyisobutyrylation sites.
• To create an accessible online tool for Khib site identification.

Main Methods:

• A deep-learning model, CNNKhib, was constructed using a convolutional neural network and one-hot encoding.
• The model was trained and validated on experimentally verified Khib sites across multiple species.
• A general model was developed by integrating data from different species.
• An online prediction tool, DeepKhib, was created incorporating both species-specific and general models.

Main Results:

• CNNKhib demonstrated superior performance compared to traditional machine-learning and other deep-learning models.
• Area under the ROC curve (AUC) values for CNNKhib ranged from 0.82 to 0.87 across different species.
• The general model achieved AUC values between 0.79 and 0.87, indicating high universality and effectiveness.
• The DeepKhib online tool provides an easy-to-use platform for Khib site prediction.

Conclusions:

• The developed deep-learning approach significantly enhances the accuracy and generalizability of lysine 2-Hydroxyisobutyrylation site prediction.
• DeepKhib offers a valuable resource for researchers studying the functional roles of Khib modifications.
• The tool facilitates further investigation into gene transcription and signal transduction pathways regulated by Khib.