GlyNet: a multi-task neural network for predicting protein-glycan interactions

Integration of RNAseq transcriptomics and N-glycomics reveal biosynthetic pathways and predict structure-specific N-glycan expression

The processes involved in protein N-glycosylation represent new therapeutic targets for diseases but their stepwise and overlapping biosynthetic processes make it challenging to identify the specific glycogenes involved. In this work, we aimed to elucidate the interactions between glycogene expression and N-glycan abundance by constructing supervised machine-learning models for each N-glycan composition. Regression models were trained to predict N-glycan abundance (response variable) from glycogene expression (predictors) using paired LC-MS/MS N-glycomic and 3'-TagSeq transcriptomic datasets from cells derived from multiple tissue origins and treatment conditions. The datasets include cells from several tissue origins - B cell, brain, colon, lung, muscle, prostate - encompassing nearly 400 N-glycan compounds and over 160 glycogenes filtered from an 18 000-gene transcriptome. Accurate models (validation R 2 > 0.8) predicted N-glycan abundance across cell types, including GLC01 (lung cancer), CCD19-Lu (lung fibroblast), and Tib-190 (B cell). Model importance scores ranked glycogene contributions to N-glycan predictions, revealing significant glycogene associations with specific N-glycan types. The predictions were consistent across input cell quantities, unlike LC-MS/MS glycomics which showed inconsistent results. This suggests that the models can reliably predict N-glycosylation even in samples with low cell amounts and by extension, single-cell samples. These findings can provide insights into cellular N-glycosylation machinery, offering potential therapeutic strategies for diseases linked to aberrant glycosylation, such as cancer, and neurodegenerative and autoimmune disorders.

Predictions from Deep Learning Propose Substantial Protein-Carbohydrate Interplay

It is a grand challenge to identify all the protein - carbohydrate interactions in an organism. Direct experiments would require extensive libraries of glycans to definitively distinguish binding from non-binding proteins. Computational screening of proteins for carbohydrate-binding provides an attractive and ultimately testable alternative. Recent computational techniques have focused primarily on which protein residues interact with carbohydrates or which carbohydrate species a protein binds to. Current estimates label 1.5 to 5% of proteins as carbohydrate-binding proteins; however, 50-70% of proteins are known to be glycosylated, suggesting a potential wealth of proteins that bind to carbohydrates. We therefore developed a novel dataset and neural network architecture, named Protein interaction of Carbohydrates Predictor (PiCAP), to predict whether a protein non-covalently binds to a carbohydrate. We trained PiCAP on a dataset of known carbohydrate binders, and we selected proteins that we identified as likely not to bind carbohydrates, including DNA-binding transcription factors, cytoskeletal components, selected antibodies, and selected small-molecule-binding proteins. PiCAP achieves a 90% balanced accuracy on protein-level predictions of carbohydrate binding/non-binding. Using the same dataset, we developed a model named Carbohydrate Protein Site Identifier 2 (CAPSIF2) to predict protein residues that interact non-covalently with carbohydrates. CAPSIF2 achieves a Dice coefficient of 0.57 on residue-level predictions on our independent test dataset, outcompeting all previous models for this task. To demonstrate the biological applicability of PiCAP and CAPSIF2, we investigated cell surface proteins of human neural cells and further predicted the likelihood of three proteomes, notably E. coli, M. musculus, and H. sapiens, to bind to carbohydrates. PiCAP predicts that approximately 35-40% of proteins in these proteomes bind carbohydrates, indicating a substantial interplay of protein-carbohydrate interactions for cellular functionality.

Tools for structural lectinomics: From structures to lectomes

Lectins are ubiquitous proteins that interact with glycans in a variety of molecular processes and as such, also play a role in diseases, whether infectious, chronic or cancer-related. The systematic study of lectins is therefore essential, in particular for understanding cell-cell communication. Accumulated protein three-dimensional structural data in the past decades boosted advance in AI-based prediction and opened up new options to characterise lectins that are known to often be multimeric and multivalent. This article reviews the methods to obtain structures of lectins, the current data available for lectin 3D structures and their interactions, how this knowledge is used to classify these proteins and shows that the combination of an array of bioinformatics tools should make the prediction of binding specificity possible in a near future.

Computational toolbox for the analysis of protein-glycan interactions

Protein-glycan interactions play pivotal roles in numerous biological processes, ranging from cellular recognition to immune response modulation. Understanding the intricate details of these interactions is crucial for deciphering the molecular mechanisms underlying various physiological and pathological conditions. Computational techniques have emerged as powerful tools that can help in drawing, building and visualising complex biomolecules and provide insights into their dynamic behaviour at atomic and molecular levels. This review provides an overview of the main computational tools useful for studying biomolecular systems, particularly glycans, both in free state and in complex with proteins, also with reference to the principles, methodologies, and applications of all-atom molecular dynamics simulations. Herein, we focused on the programs that are generally employed for preparing protein and glycan input files to execute molecular dynamics simulations and analyse the corresponding results. The presented computational toolbox represents a valuable resource for researchers studying protein-glycan interactions and incorporates advanced computational methods for building, visualising and predicting protein/glycan structures, modelling protein-ligand complexes, and analyse MD outcomes. Moreover, selected case studies have been reported to highlight the importance of computational tools in studying protein-glycan systems, revealing the capability of these tools to provide valuable insights into the binding kinetics, energetics, and structural determinants that govern specific molecular interactions.

LeGenD: determining N-glycoprofiles using an explainable AI-leveraged model with lectin profiling

Glycosylation affects many vital functions of organisms. Therefore, its surveillance is critical from basic science to biotechnology, including biopharmaceutical development and clinical diagnostics. However, conventional glycan structure analysis faces challenges with throughput and cost. Lectins offer an alternative approach for analyzing glycans, but they only provide glycan epitopes and not full glycan structure information. To overcome these limitations, we developed LeGenD, a lectin and AI-based approach to predict N-glycan structures and determine their relative abundance in purified proteins based on lectin-binding patterns. We trained the LeGenD model using 309 glycoprofiles from 10 recombinant proteins, produced in 30 glycoengineered CHO cell lines. Our approach accurately reconstructed experimentally-measured N-glycoprofiles of bovine Fetuin B and IgG from human sera. Explanatory AI analysis with SHapley Additive exPlanations (SHAP) helped identify the critical lectins for glycoprofile predictions. Our LeGenD approach thus presents an alternative approach for N-glycan analysis.

HumanLectome, an update of UniLectin for the annotation and prediction of human lectins

The UniLectin portal (https://unilectin.unige.ch/) was designed in 2019 with the goal of centralising curated and predicted data on carbohydrate-binding proteins known as lectins. UniLectin is also intended as a support for the study of lectomes (full lectin set) of organisms or tissues. The present update describes the inclusion of several new modules and details the latest (https://unilectin.unige.ch/humanLectome/), covering our knowledge of the human lectome and comprising 215 unevenly characterised lectins, particularly in terms of structural information. Each HumanLectome entry is protein-centric and compiles evidence of carbohydrate recognition domain(s), specificity, 3D-structure, tissue-based expression and related genomic data. Other recent improvements regarding interoperability and accessibility are outlined.

Structure-based neural network protein-carbohydrate interaction predictions at the residue level

Carbohydrates dynamically and transiently interact with proteins for cell-cell recognition, cellular differentiation, immune response, and many other cellular processes. Despite the molecular importance of these interactions, there are currently few reliable computational tools to predict potential carbohydrate-binding sites on any given protein. Here, we present two deep learning (DL) models named CArbohydrate-Protein interaction Site IdentiFier (CAPSIF) that predicts non-covalent carbohydrate-binding sites on proteins: (1) a 3D-UNet voxel-based neural network model (CAPSIF:V) and (2) an equivariant graph neural network model (CAPSIF:G). While both models outperform previous surrogate methods used for carbohydrate-binding site prediction, CAPSIF:V performs better than CAPSIF:G, achieving test Dice scores of 0.597 and 0.543 and test set Matthews correlation coefficients (MCCs) of 0.599 and 0.538, respectively. We further tested CAPSIF:V on AlphaFold2-predicted protein structures. CAPSIF:V performed equivalently on both experimentally determined structures and AlphaFold2-predicted structures. Finally, we demonstrate how CAPSIF models can be used in conjunction with local glycan-docking protocols, such as GlycanDock, to predict bound protein-carbohydrate structures.

Structure-Based Neural Network Protein-Carbohydrate Interaction Predictions at the Residue Level

Carbohydrates dynamically and transiently interact with proteins for cell-cell recognition, cellular differentiation, immune response, and many other cellular processes. Despite the molecular importance of these interactions, there are currently few reliable computational tools to predict potential carbohydrate binding sites on any given protein. Here, we present two deep learning models named CArbohydrate-Protein interaction Site IdentiFier (CAPSIF) that predict carbohydrate binding sites on proteins: (1) a 3D-UNet voxel-based neural network model (CAPSIF:V) and (2) an equivariant graph neural network model (CAPSIF:G). While both models outperform previous surrogate methods used for carbohydrate binding site prediction, CAPSIF:V performs better than CAPSIF:G, achieving test Dice scores of 0.597 and 0.543 and test set Matthews correlation coefficients (MCCs) of 0.599 and 0.538, respectively. We further tested CAPSIF:V on AlphaFold2-predicted protein structures. CAPSIF:V performed equivalently on both experimentally determined structures and AlphaFold2 predicted structures. Finally, we demonstrate how CAPSIF models can be used in conjunction with local glycan-docking protocols, such as GlycanDock, to predict bound protein-carbohydrate structures.

Glycoinformatics in the Artificial Intelligence Era

Artificial intelligence (AI) methods have been and are now being increasingly integrated in prediction software implemented in bioinformatics and its glycoscience branch known as glycoinformatics. AI techniques have evolved in the past decades, and their applications in glycoscience are not yet widespread. This limited use is partly explained by the peculiarities of glyco-data that are notoriously hard to produce and analyze. Nonetheless, as time goes, the accumulation of glycomics, glycoproteomics, and glycan-binding data has reached a point where even the most recent deep learning methods can provide predictors with good performance. We discuss the historical development of the application of various AI methods in the broader field of glycoinformatics. A particular focus is placed on shining a light on challenges in glyco-data handling, contextualized by lessons learnt from related disciplines. Ending on the discussion of state-of-the-art deep learning approaches in glycoinformatics, we also envision the future of glycoinformatics, including development that need to occur in order to truly unleash the capabilities of glycoscience in the systems biology era.

An analytical study on the identification of N-linked glycosylation sites using machine learning model

N-linked is the most common type of glycosylation which plays a significant role in identifying various diseases such as type I diabetes and cancer and helps in drug development. Most of the proteins cannot perform their biological and psychological functionalities without undergoing such modification. Therefore, it is essential to identify such sites by computational techniques because of experimental limitations. This study aims to analyze and synthesize the progress to discover N-linked places using machine learning methods. It also explores the performance of currently available tools to predict such sites. Almost seventy research articles published in recognized journals of the N-linked glycosylation field have shortlisted after the rigorous filtering process. The findings of the studies have been reported based on multiple aspects: publication channel, feature set construction method, training algorithm, and performance evaluation. Moreover, a literature survey has developed a taxonomy of N-linked sequence identification. Our study focuses on the performance evaluation criteria, and the importance of N-linked glycosylation motivates us to discover resources that use computational methods instead of the experimental method due to its limitations.