Abstract:Lysine crotonylation (Kcr) is an evolutionarily conserved protein post-translational modifications, which plays an important role in cellular physiology and pathology, such as chromatin remodeling, gene transcription regulation, telomere maintenance, inflammation, and cancer. Tandem mass spectrometry (LC-MS/MS) has been used to identify the global Kcr profiling of human, at the same time, many computing methods have been developed to predict Kcr sites without high experiment cost. Deep learning network solves the problem of manual feature design and selection in traditional machine learning (NLP), especially the algorithms in natural language processing which treated peptides as sentences, thus can extract more in-depth information and obtain higher accuracy. In this work, we establish a Kcr prediction model named ATCLSTM-Kcr which use self-attention mechanism combined with NLP method to highlight the important features and further capture the internal correlation of the features, to realize the feature enhancement and noise reduction modules of the model. Independent tests have proved that ATCLSTM-Kcr has better accuracy and robustness than similar prediction tools. Then, we design pipeline to generate MS-based benchmark dataset to avoid the false negatives caused by MS-detectability and improve the sensitivity of Kcr prediction. Finally, we develop a Human Lysine Crotonylation Database (HLCD) which using ATCLSTM-Kcr and the two representative deep learning models to score all lysine sites of human proteome, and annotate all Kcr sites identified by MS of current published literatures. HLCD provides an integrated platform for human Kcr sites prediction and screening through multiple prediction scores and conditions, and can be accessed on the website:www.urimarker.com/HLCD/. SIGNIFICANCE: Lysine crotonylation (Kcr) plays an important role in cellular physiology and pathology, such as chromatin remodeling, gene transcription regulation and cancer. To better elucidate the molecular mechanisms of crotonylation and reduce the high experimental cost, we establish a deep learning Kcr prediction model and solve the problem of false negatives caused by the detectability of mass spectrometry (MS). Finally, we develop a Human Lysine Crotonylation Database to score all lysine sites of human proteome, and annotate all Kcr sites identified by MS of current published literatures. Our work provides a convenient platform for human Kcr sites prediction and screening through multiple prediction scores and conditions.

BERT-Kcr: prediction of lysine crotonylation sites by a transfer learning method with pre-trained BERT models

Identification of Protein Lysine Crotonylation Sites by a Deep Learning Framework with Convolutional Neural Networks

Deep-Kcr: accurate detection of lysine crotonylation sites using deep learning method

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

MSTL-Kace: Prediction of Prokaryotic Lysine Acetylation Sites Based on Multistage Transfer Learning Strategy

nhKcr: a new bioinformatics tool for predicting crotonylation sites on human nonhistone proteins based on deep learning

Adapt-Kcr: a Novel Deep Learning Framework for Accurate Prediction of Lysine Crotonylation Sites Based on Learning Embedding Features and Attention Architecture.

Using ATCLSTM-Kcr to predict and generate the human lysine crotonylation database

DeepCap-Kcr: Accurate Identification and Investigation of Protein Lysine Crotonylation Sites Based on Capsule Network.

Large-scale comparative assessment of computational predictors for lysine post-translational modification sites

Auto-Kla: a novel web server to discriminate lysine lactylation sites using automated machine learning

Ikcr_cnn: A Novel Computational Tool for Imbalance Classification of Human Nonhistone Crotonylation Sites Based on Convolutional Neural Networks with Focal Loss.

ILYCROsite: Identification of lysine crotonylation sites based on FCM-GRNN undersampling technique

Prediction of protein N-terminal acetylation modification sites based on CNN-BiLSTM-attention model

PlantNh-Kcr: a deep learning model for predicting non-histone crotonylation sites in plants

DeepKhib: A Deep-Learning Framework for Lysine 2-Hydroxyisobutyrylation Sites Prediction

Prediction and Functional Analysis of Prokaryote Lysine Acetylation Site by Incorporating Six Types of Features into Chou's General PseAAC

SLAM: Structure-aware lysine β-hydroxybutyrylation prediction with protein language model

predML-Site: Predicting Multiple Lysine PTM Sites With Optimal Feature Representation and Data Imbalance Minimization

KbhbXG: A Machine Learning Architecture Based on XGBoost for Prediction of Lysine Β-Hydroxybutyrylation (kbhb) Modification Sites

Improved Species-Specific Lysine Acetylation Site Prediction Based On A Large Variety Of Features Set