Recent Development of Machine Learning Methods in Microbial Phosphorylation Sites

A Review of Machine Learning and Algorithmic Methods for Protein Phosphorylation Site Prediction

Post-translational modifications (PTMs) have key roles in extending the functional diversity of proteins and, as a result, regulating diverse cellular processes in prokaryotic and eukaryotic organisms. Phosphorylation modification is a vital PTM that occurs in most proteins and plays a significant role in many biological processes. Disorders in the phosphorylation process lead to multiple diseases, including neurological disorders and cancers. The purpose of this review is to organize this body of knowledge associated with phosphorylation site (p-site) prediction to facilitate future research in this field. At first, we comprehensively review all related databases and introduce all steps regarding dataset creation, data preprocessing, and method evaluation in p-site prediction. Next, we investigate p-site prediction methods, which are divided into two computational groups: algorithmic and machine learning (ML). Additionally, it is shown that there are basically two main approaches for p-site prediction by ML: conventional and end-to-end deep learning methods, both of which are given an overview. Moreover, this review introduces the most important feature extraction techniques, which have mostly been used in p-site prediction. Finally, we create three test sets from new proteins related to the released version of the database of protein post-translational modifications (dbPTM) in 2022 based on general and human species. Evaluating online p-site prediction tools on newly added proteins introduced in the dbPTM 2022 release, distinct from those in the dbPTM 2019 release, reveals their limitations. In other words, the actual performance of these online p-site prediction tools on unseen proteins is notably lower than the results reported in their respective research papers.

Identification of phosphorylation site using S-padding strategy based convolutional neural network

Purpose

Abnormal phosphorylation has been proved to associate with a variety of human diseases, and the identification of phosphorylation sites is one of the research hotspots in healthcare. The study of phosphorylation site prediction in deep learning models often introduces a variety of information, and the utilization of complex models limits the usage scenarios of the models.

Methods

An enhanced deep learning method with S-padding strategy based on convolutional neural network is proposed in this paper. The S-padding strategy forms a three-dimensional matrix with extension information from original amino acid sequences, and a corresponding 2D-CNN model is designed to abstract the comprehensive features of phosphorylation site area in protein sequences.

Results

The fivefold cross-validation experiments are conducted, and the results show the performance of the proposed method on human dataset can achieve an accuracy of 89.68 % on serine/threonine sites and 88.16 % on tyrosine sites, respectively. Furthermore, phosphorylation site prediction on different organisms obtains the accuracy, sensitivity, and specificity of over 0.85, indicating a potential capability on phosphorylation site prediction task. Comparison result with existing models shows that the proposed method obtains better performance on both accuracy and AUC value, and the proposed method can further improve performance with sufficient training data.

Conclusion

This method enables proteome-wide predictions via models trained on a large amount of phosphorylation data, further exploiting the potential of protein phosphorylation site identification, and helping to provide insights into phosphorylation mechanisms.

SubFeat: Feature subspacing ensemble classifier for function prediction of DNA, RNA and protein sequences

The information of a cell is primarily contained in deoxyribonucleic acid (DNA). There is a flow of DNA information to protein sequences via ribonucleic acids (RNA) through transcription and translation. These entities are vital for the genetic process. Recent epigenetics developments also show the importance of the genetic material and knowledge of their attributes and functions. However, the growth in these entities' available features or functionalities is still slow due to the time-consuming and expensive in vitro experimental methods. In this paper, we have proposed an ensemble classification algorithm called SubFeat to predict biological entities' functionalities from different types of datasets. Our model uses a feature subspace-based novel ensemble method. It divides the feature space into sub-spaces, which are then passed to learn individual classifier models. The ensemble is built on these base classifiers that use a weighted majority voting mechanism. SubFeat tested on four datasets comprising two DNA, one RNA, and one protein dataset, and it outperformed all the existing single classifiers and the ensemble classifiers. SubFeat is made available as a Python-based tool. We have made the package SubFeat available online along with a user manual. It is freely accessible from here: https://github.com/fazlulhaquejony/SubFeat.

Improved Prediction of Protein-Protein Interaction Mapping on Homo Sapiens by Using Amino Acid Sequence Features in a Supervised Learning Framework

BACKGROUND

Protein-Protein Interaction (PPI) has emerged as a key role in the control of many biological processes including protein function, disease incidence, and therapy design. However, the identification of PPI by wet lab experiment is a challenging task, since it is laborious, time consuming and expensive. Therefore, computational prediction of PPI is now given emphasis before going to the experimental validation, since it is simultaneously less laborious, time saver and cost minimizer.

OBJECTIVE

The objective of this study is to develop an improved computational method for PPI prediction mapping on Homo sapiens by using the amino acid sequence features in a supervised learning framework.

METHODS

The experimentally validated 91 positive-PPI pairs of human protein sequences were collected from IntAct Molecular Interaction Database. Then we constructed three balanced datasets with ratios 1:1, 1:2 and 1:3 of positive and negative PPI samples. Then we partitioned each dataset into training (80%) and independent test (20%) datasets. Again each training dataset was partitioned into four mutually exclusive groups of equal sizes for interchanging each group with independent test group to perform 5-fold cross validation (CV). Then we trained candidate seven classifiers (NN, SVM, LR, NB, KNN, AB and RF) with each ratio case to obtain the better PPI predictor by comparing their performance scores.

RESULTS

The random forest (RF) based predictor that was trained with 1:2 ratio of positive-PPI and negative-PPI samples based on AAC encoding features provided the most accurate PPI prediction by producing the highest average performance scores of accuracy (93.50%), sensitivity (95.0%), MCC (85.2%), AUC (0.941) and pAUC (0.236) with the 5-fold cross-validation. It also achieved the highest average performance scores of accuracy (92.0%), sensitivity (94.0%), MCC (83.6%), AUC (0.922) and pAUC (0.207) with the independent test datasets in a comparison of the other candidate and existing predictors.

CONCLUSION

The final resultant prediction strongly recommend that the RF based predictor is a better prediction model of PPI mapping on Homo sapiens.

An Improved Computational Prediction Model for Lysine Succinylation Sites Mapping on Homo sapiens by Fusing Three Sequence Encoding Schemes with the Random Forest Classifier

Background

Lysine succinylation is one of the reversible protein post-translational modifications (PTMs), which regulate the structure and function of proteins. It plays a significant role in various cellular physiologies including some diseases of human as well as many other organisms. The accurate identification of succinylation site is essential to understand the various biological functions and drug development.

Methods

In this study, we developed an improved method to predict lysine succinylation sites mapping on Homo sapiens by the fusion of three encoding schemes such as binary, the composition of k-spaced amino acid pairs (CKSAAP) and amino acid composition (AAC) with the random forest (RF) classifier. The prediction performance of the proposed random forest (RF) based on the fusion model in a comparison of other candidates was investigated by using 20-fold cross-validation (CV) and two independent test datasets were collected from two different sources.

Results

The CV results showed that the proposed predictor achieves the highest scores of sensitivity (SN) as 0.800, specificity (SP) as 0.902, accuracy (ACC) as 0.919, Mathew correlation coefficient (MCC) as 0.766 and partial AUC (pAUC) as 0.163 at a false-positive rate (FPR) = 0.10 and area under the ROC curve (AUC) as 0.958. It achieved the highest performance scores of SN as 0.811, SP as 0.902, ACC as 0.891, MCC as 0.629 and pAUC as 0.139 and AUC as 0.921 for the independent test protein set-1 and SN as 0.772, SP as 0.901, ACC as 0.836, MCC as 0.677 and pAUC as 0.141 at FPR = 0.10 and AUC as 0.923 for the independent test protein set-2. It also outperformed all the other existing prediction models.

Conclusion

The prediction performances as discussed in this article recommend that the proposed method might be a useful and encouraging computational resource for lysine succinylation site prediction in the case of human population.