Protein ubiquitination occurs when the ubiquitin protein binds to a target protein residue of lysine (K), and it is an important regulator of many cellular functions, such as signal transduction, cell division, and immune reactions, in eukaryotes. Experimental and clinical studies have shown that ubiquitination plays a key role in several human diseases, and recent advances in proteomic technology have spurred interest in identifying ubiquitination sites. However, most current computing tools for predicting target sites are based on small-scale data and shallow machine learning algorithms. As more experimentally validated ubiquitination sites emerge, we need to design a predictor that can identify lysine ubiquitination sites in large-scale proteome data. In this work, we propose a deep learning predictor, DeepUbi, based on convolutional neural networks. Four different features are adopted from the sequences and physicochemical properties. In a 10-fold cross validation, DeepUbi obtains an AUC (area under the Receiver Operating Characteristic curve) of 0.9, and the accuracy, sensitivity and specificity exceeded 85%. The more comprehensive indicator, MCC, reaches 0.78. We also develop a software package that can be freely downloaded from https://github.com/Sunmile/DeepUbi . Our results show that DeepUbi has excellent performance in predicting ubiquitination based on large data.

Lysine acetylation in protein is one of the most important post-translational modifications (PTMs). It plays an important role in essential biological processes and is related to various diseases. To obtain a comprehensive understanding of regulatory mechanism of lysine acetylation, the key is to identify lysine acetylation sites. Previously, several shallow machine learning algorithms had been applied to predict lysine modification sites in proteins. However, shallow machine learning has some disadvantages. For instance, it is not as effective as deep learning for processing big data. In this work, a novel predictor named DeepAcet was developed to predict acetylation sites. Six encoding schemes were adopted, including a one-hot, BLOSUM62 matrix, a composition of K-space amino acid pairs, information gain, physicochemical properties, and a position specific scoring matrix to represent the modified residues. A multilayer perceptron (MLP) was utilized to construct a model to predict lysine acetylation sites in proteins with many different features. We also integrated all features and implemented the feature selection method to select a feature set that contained 2199 features. As a result, the best prediction achieved 84.95% accuracy, 83.45% specificity, 86.44% sensitivity, 0.8540 AUC, and 0.6993 MCC in a 10-fold cross-validation. For an independent test set, the prediction achieved 84.87% accuracy, 83.46% specificity, 86.28% sensitivity, 0.8407 AUC, and 0.6977 MCC. The predictive performance of our DeepAcet is better than that of other existing methods. DeepAcet can be freely downloaded from https://github.com/Sunmile/DeepAcet .

Clonal evolution drives cancer progression and therapeutic resistance. Recent studies have revealed divergent longitudinal trajectories in gliomas, but early molecular features steering posttreatment cancer evolution remain unclear. Here, we collected sequencing and clinical data of initial-recurrent tumor pairs from 544 adult diffuse gliomas and performed multivariate analysis to identify early molecular predictors of tumor evolution in three diffuse glioma subtypes. We found that CDKN2A deletion at initial diagnosis preceded tumor necrosis and microvascular proliferation that occur at later stages of IDH-mutant glioma. Ki67 expression at diagnosis was positively correlated with acquiring hypermutation at recurrence in the IDH–wild-type glioma. In all glioma subtypes, MYC gain or MYC- target activation at diagnosis was associated with treatment-induced hypermutation at recurrence. To predict glioma evolution, we constructed CELLO2 (Cancer EvoLution for LOngitudinal data version 2), a machine learning model integrating features at diagnosis to forecast hypermutation and progression after treatment. CELLO2 successfully stratified patients into subgroups with distinct prognoses and identified a high-risk patient group featured by MYC gain with worse post-progression survival, from the low-grade IDH-mutant-noncodel subtype. We then performed chronic temozolomide-induction experiments in glioma cell lines and isogenic patient-derived gliomaspheres and demonstrated that MYC drives temozolomide resistance by promoting hypermutation. Mechanistically, we demonstrated that, by binding to open chromatin and transcriptionally active genomic regions, c-MYC increases the vulnerability of key mismatch repair genes to treatment-induced mutagenesis, thus triggering hypermutation. This study reveals early predictors of cancer evolution under therapy and provides a resource for precision oncology targeting cancer dynamics in diffuse gliomas.

As one of the new posttranslational modification, lysine glutarylation has been identified in both prokaryotic and eukaryotic cells. These glutarylated proteins are involved in various cellular functions, such as translation, metabolism, and exhibited diverse subcellular localizations. Experimental identification of lysine glutarylation sites was founded in 2014 and also identified its deglutarylase sirturn 5(SIRT 5). Computational prediction of lysine glutarylation could be a complementary way to the experimental technique. In this work, the lysine glutarylation predictor iGlu-Lys has been developed based on the machine learning scheme. We have selected the best feature scheme which took the amino acid pair order and special-position information into account from four constructions. The machine learning algorithm support vector machine has been adopted and its performance has been measured for different window length of peptides. In the 10-fold cross-validation with window length 19, the AUC and MCC were 0.8944 and 0.5098, respectively. Different ROC curves in 6-, 8-, and 10-fold cross-validations were very close which illustrated the robustness of our predictor. The results of iGLu-Lys were better than the existing method GlutPred. Meanwhile, a free webserver for iGlu-Lys is accessible at http://app.aporc.org/iGlu-Lys/.