Computational Prediction of Protein Post-Translational Modification Sites using ML Algorithms

After a protein has been translated by a ribosome, it may go through a processknown as post-translational modification (PTM). PTM is an alteration to one or moreamino acids in a protein that happens after it has been translated. They are also regarded as a significant component of cell signaling and networking pathways. Methylation is often considered to be one of the most significant forms of PTMs. Methylationis playing a crucial impact in the maintenance of the stability, dynamic equilibrium,and remodeling of chromatins and causes certain abnormalities in cells towards severeillnesses. Identifying methylation in the presence of a methylation marker can be doneusing methylation-specific antibodies, mass spectrometry, or radioactive labeling.However, traditional procedures required additional expenses and time-consumingspecialized equipment. In this study, we demonstrate the use of sequential-basedfeatures, which are computationally less expensive to produce. We developed a computational model for the prediction of methylation PTM sites that incorporates bothevolutionary sequence-based features and structural-based information. We also useda SMOTETomek-based hybrid sampling method at particular ratios in conjunctionwith features based on mutual information. Following that, The optimal collection offeatures including permutation and Gini feature importance was applied to multiplealgorithms to predict the result, and the final model was trained using an extremegradient boosting (XGBoost) classifier. With 10-fold cross-validation and an independent test set, our model achieves an AUC-ROC score of 0.94 and 0.90 for lysinemethylation sites, demonstrating promising performance. Our proposed techniquesignificantly outperforms and beats the independent models by a wide margin. Weanticipate that the higher performance of our model will benefit researchers involvedin the identification of PTM methylation sites. Our model is easily accessible for useas a standalone predictor at https://github.com/arafatro/KmethPred.