Deep Learning Techniques Applied to Phosphorylation Site Prediction: A Systematic Review

This paper presents a systematic literature review (SLR) investigating the use of deep learning techniques for predicting phosphorylation sites on proteins. Post-translational modifications, especially phosphorylation, play a key role in regulating various cellular processes, and accurate identification of phosphorylation sites is crucial for understanding diseases and developing therapeutic strategies. Given the limitations of conventional experimental methods, deep learning has emerged as a promising tool for large-scale and efficient prediction. The SLR process involved a search of multiple databases, resulting in the rigorous selection of 24 papers that were analyzed based on general information, databases, and metrics. The review found that deep learning techniques have shown promising results, with convolutional neural networks (CNN) being the most commonly used technique and CNN + long short-term memory demonstrating the highest average performance in terms of area under the curve. Most models focused on amino acid sequences centered around serine, threonine, or tyrosine residues and one on histidine, with datasets from multiple organisms. The SLR also identified gaps in the literature, such as the need to explore transformer-based models like BERT and the phosphorylation of other amino acids, including arginine, lysine, aspartate, glutamate, and cysteine.