A Novel Deep Learning Approach for the Prediction of Arabidopsis Thaliana Ubiquitination Sites

Abstract

Protein ubiquitination is a crucial post-translational modification involving the attachment of ubiquitin molecules to proteins, forming ubiquitin-protein complexes. This modification plays a pivotal role in various biological processes, including protein decomposition, regulation of enzymatic activity, modulation of interactions, cell cycle regulation, and the onset of serious diseases such as cancer, diabetes, Parkinson's, Alzheimer's, and cardiovascular diseases. Scientists have devoted extensive research to developing tools for predicting ubiquitination in different species. These tools primarily rely on predefined sequence features and machine learning algorithms. However, the variations in the ubiquitination cascade among species remain poorly understood. While machine learning algorithms typically focus on the physical and chemical characteristics of previously analyzed proteins, deep learning algorithms can automatically extract features from protein language strings. Nevertheless, there are currently limited studies that simultaneously incorporate both types of features. In this study, we present a novel approach for predicting ubiquitination sites in Arabidopsis thaliana. We build a deep learning-based combined model that integrates both the chemical and physical characteristics of proteins and other natural language features of proteins. Our results demonstrate that our proposed model outperforms previous machine learning algorithms and prediction tools for A. thaliana ubiquitination sites. We anticipate that these findings will prove valuable to researchers in their respective studies.