Prediction and analysis of multiple protein lysine modified sites based on conditional wasserstein generative adversarial networks

Abstract

Background Protein post-translational modification (PTM) is a key issue to investigate the mechanism of protein’s function. With the rapid development of proteomics technology, a large amount of protein sequence data has been generated, which highlights the importance of the in-depth study and analysis of PTMs in proteins. Method We proposed a new multi-classification machine learning pipeline MultiLyGAN to identity seven types of lysine modified sites. Using eight different sequential and five structural construction methods, 1497 valid features were remained after the filtering by Pearson correlation coefficient. To solve the data imbalance problem, Conditional Generative Adversarial Network (CGAN) and Conditional Wasserstein Generative Adversarial Network (CWGAN), two influential deep generative methods were leveraged and compared to generate new samples for the types with fewer samples. Finally, random forest algorithm was utilized to predict seven categories. Results In the tenfold cross-validation, accuracy (Acc) and Matthews correlation coefficient (MCC) were 0.8589 and 0.8376, respectively. In the independent test, Acc and MCC were 0.8549 and 0.8330, respectively. The results indicated that CWGAN better solved the existing data imbalance and stabilized the training error. Alternatively, an accumulated feature importance analysis reported that CKSAAP, PWM and structural features were the three most important feature-encoding schemes. MultiLyGAN can be found at https://github.com/Lab-Xu/MultiLyGAN. Conclusions The CWGAN greatly improved the predictive performance in all experiments. Features derived from CKSAAP, PWM and structure schemes are the most informative and had the greatest contribution to the prediction of PTM.

VirusHound-I: prediction of viral proteins involved in the evasion of host adaptive immune response using the random forest algorithm and generative adversarial network for data augmentation

“…Within the field of deep learning, GANs have been shown to be very useful in the development of bioinformatic predictive models [ 44 ]. For example, this type of neural network has been used as a data augmentation technique in the study of protein post-translational modifications [ 45 ], antiviral peptides [ 28 ] and protein solubility [ 46 ], among other cases reported in an excellent review by Wan et al . [ 44 ], showing excellent results.…”

BERT-Kgly: A Bidirectional Encoder Representations From Transformers (BERT)-Based Model for Predicting Lysine Glycation Site for Homo sapiens

“…However, wet experiment methods are both cost- and time-consuming. Alternatively, several in silico methods ( Johansen et al, 2006 ; Liu et al, 2015 ; Ju et al, 2017 ; Xu H. et al, 2017 ; Zhao et al, 2017 ; Islam et al, 2018 ; Chen K. et al, 2019 ; Yu et al, 2019 ; Khanum et al, 2020 ; Yang et al, 2021 ; Yao et al, 2021 ) have been developed to predict the Kgly sites efficiently. In a pioneer work, Johansen et al proposed a predictor, GlyNN, built by neural networks based on a dataset with 89 Kgly sites and 126 non-Kgly sites of 20 proteins ( Johansen et al, 2006 ).…”

“…Leveraging Xu et al’s dataset as training dataset, Yao et al developed a model, ABC-Gly, by selecting the optimal feature subset with a two-step feature selection method by combining the Fisher score and an improved binary artificial bee colony algorithm ( Yao et al, 2021 ). All the previous methods were built on the dataset with less than 500 Kgly sites; however, four other methods, PredGly ( Yu et al, 2019 ), Gly-LysPred ( Khanum et al, 2020 ), MUscADEL ( Chen Z et al, 2019 ), and MultiLyGAN ( Yang et al, 2021 ), which were built on datasets with more than 1,000 Kgly sites. For building PredGly, Yu et al (2019) collected Kgly sites from PLMD ( Xu H. et al, 2017 ) and used CD-HIT ( Huang et al, 2010 ) to remove the redundancy for protein sequences and peptide segments, with a cutoff of 30%.…”

“…By collecting Kgly sites from UniProt ( https://www.uniprot.org/ ), Khanum et al obtained their dataset with 1,287 Kgly sites and 1,300 non-Kgly sites by using CD-HIT to remove the redundancy, with a cutoff of 60%, and then built their model by using random forest ( Khanum et al, 2020 ). Both MUscADEL ( Chen K et al, 2019 ) and MultiLyGAN ( Yang et al, 2021 ) were developed to predict multiple lysine modification sites. For MUscADEL, Chen et al collected Kgly sites for both Homo sapiens and Mus musculus from the PhosphoSitePlus database ( Hornbeck et al, 2015 ), and then removed the redundancy of protein sequences by using CD-HIT ( Huang et al, 2010 ), with a cutoff of 30%.…”

“…Based on the dataset with 3,209 Kgly sites, they built their model by using a deep learning algorithm. In MultiLyGAN ( Yang et al, 2021 ), Yang et al collected lysine modification sites from the CPLM2.0 database ( Liu et al, 2014 ), and after removing redundancy by using CD-HIT at the segment level with a cutoff of 40%, they obtained 1,454 Kgly sites. Their model is a multiple-label model built with data augmentation by conditional Wasserstein generative adversarial networks.…”