Protein carbonylation sites prediction by using deep neural networks
Loading...
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UMT, Lahore
Abstract
Post-translational modifications (PTMs) are the processing events that can modify the properties of a protein in case of proteolysis or if any modifying group is added to one or more amino acids. Oxidative stress arises when there is an imbalance in the regulation and production of reactive oxygen species (ROS) and reactive nitrogen species (RNS). The protein carbonylation is a post-translational modification and a biomarker foroxidative stress because of some special traits it has such as its early development, irreversibility, and stability. Overabundance in external oxidative stress, obesity, and aging increase the density of protein carbonylation, and this enlargement signals early-stage diseases. There are many human diseases associated with protein carbonylation such as diabetes, chronic renal failure, sepsis, chronic lung disease, Parkinson’s disease, and Alzheimer’s disease. So, it is very important to identify protein carbonylation sites in pathology and physiology. It can provide important evidence for basic research and medication development. Carbonylation is vulnerable to only a subset of proteins and most of the carbonyl groups are made from four types of amino acid residues known as lysine (K), proline (P), arginine (R), and threonine (T). It is very costly and time-consuming to determine the carbonylation sites experimentally, especially in the case of broad datasets. So, computational methods are recommended for the recognition of carbonylation sites in proteins. This research develops a protein carbonylation sites predictor. An experimentally verified benchmark dataset is used for the study. Three deep neural networks are trained and tested; fully connected neural network, recurrent neural network, and convolutional neural network. Recurrent Neural Networks (RNNs) are trained with simple RNN units, Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) units. The convolutional neural network provided the best performance with an accuracy value of 0.913. So, it can learn deep representations among data efficiently. This study has the novelty to use deep neural networks for carbonylation sites prediction. It is providing comparable results with state-of-the-art predictors of carbonylation sites. Only two models are providing better performance than the proposed model. I have not done any feature extraction and still, I got comparable results. All of the other previously proposed predictors have low performance than the proposed methodology. So, the proposed model is stable, valid, and reliable than all of these predictors. This study is providing a baseline to researchers who want to work with deep neural networks for carbonylation sites prediction.