Identification of the ubiquitin-proteasome pathway domain by hyperparameter optimization based on a 2D convolutional...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• The ubiquitin-proteasome pathway (UPP) is crucial for cellular protein degradation, impacting numerous biological processes.
• Dysregulation of the UPP is implicated in various human diseases, highlighting the need for better understanding and analysis tools.
• The increasing volume of unannotated sequencing data necessitates efficient computational methods for UPP identification.

Purpose of the Study:

• To develop a novel deep learning-based predictor, 2DCNN-UPP, for accurate identification of UPP proteins.
• To address the challenges posed by large-scale, unannotated biological data in UPP characterization.
• To provide a computational tool that aids in understanding cellular processes and facilitates drug discovery related to the UPP.

Main Methods:

• Utilized a two-dimensional convolutional neural network (2D-CNN) architecture.
• Incorporated dipeptide deviation features and employed a genetic algorithm for optimal feature selection to prevent overfitting.
• Trained the model using a 10-fold cross-validation approach and evaluated its performance on an independent test set.

Main Results:

• The 2DCNN-UPP predictor achieved high accuracy, sensitivity, specificity, and an AUC (ROC) value of 0.862.
• Demonstrated superior performance compared to existing state-of-the-art methods in classifying UPP-related proteins.
• The model showed strong generalization capabilities on an independent test set, with an accuracy of 0.862 and MCC of 0.730.

Conclusions:

• 2DCNN-UPP is an effective computational tool for the rapid and accurate identification of UPP proteins.
• The developed method can aid in analyzing large-scale protein relationships and advance computational biological research.
• The approach, utilizing Dipeptide Deviation from Expected Mean (DDE) features, has potential applications in predicting protein structure, functions, and interactions with other molecules.