Biology subjects

Gao, C.

Publications and source records attributed to Gao, C..

At least 127 records · Page 7Linked to original sources

bioRxiv2022

Osteocytes regulate organismal senescence of bone and bone marrow

The skeletal system contains a series of sophisticated cellular lineages arisen from the mesenchymal stem cells (MSC) and hematopoietic stem cells (HSC), that determine the homeostasis of bone and bone marrow. Here we reasoned that osteocyte may exert a function in regulation of these lineage cell specifications and tissue homeostasis. Using a mouse model of conditional deletion of osteocytes by the expression of diphtheria toxin subunit (DTA) in dentin matrix protein 1 (DMP-1) positive osteocytes, we demonstrated that partial ablation of DMP-1positive osteocytes caused severe sarcopenia, osteoporosis and degenerative kyphosis, leading to shorter lifespan in these animals. Osteocyte reduction altered mesenchymal lineage commitment resulting in impairment of osteogenesis and induction of osteoclastogensis. Single cell RNA sequencing further revealed that hematopoietic lineage was mobilized towards myeloid lineage differentiation with expanded myeloid progenitors, neutrophils and monocytes, while the lymphopoiesis was impaired with reduced B cells in the osteocyte ablation mice. The acquisition of a senescence-associated secretory phenotype (SASP) in both osteoprogenic and myeloid lineage cells was the underlying cause. Together, we showed that osteocytes play critical roles in regulating of lineage cell specifications in bone and bone marrow through mediation of organismal senescence.

cell biology↗

bioRxiv2022

Audiovisual Integration in the Human Brain: A Coordinate-based Meta-analysis

People can seamlessly integrate a vast array of information from what we see and hear in the noisy and uncertain world. However, the neural underpinnings of audiovisual integration continue to be the topic of debate. Using strict inclusion criteria, we performed activation likelihood estimation meta-analysis on 121 neuroimaging experiments with a total of 2,092 participants. We found that audiovisual integration is linked with the coexistence of multiple integration sites including early cortical, subcortical, and higher association areas. Although activity was consistently found within the superior temporal cortex, different portions of this cortical region were identified depending on the analytical contrast used, complexity of the stimuli, and modality within which attention was directed. The context-dependent neural activity related to audiovisual integration suggests a flexible rather than fixed neural pathway for audiovisual integration. Together, our findings highlight a flexible multiple pathways model for audiovisual integration, with superior temporal cortex as the central node in these neural assemblies.

neuroscience↗

bioRxiv2022

The Spike protein of SARS-CoV-2 impairs lipid metabolism and increases susceptibility to lipotoxicity: implication for a role of Nrf2

Background/objectivesCoronavirus disease 2019 (COVID-19) patients exhibit lipid metabolic alterations, but the mechanism remains unknown. In this study, we aimed to investigate whether the Spike protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) impairs lipid metabolism in host cells. MethodsA Spike cell line in HEK293 was generated using the pcDNA vector carrying the Spike gene expression cassette. A control cell line was generated using the empty pcDNA vector. Gene expression profiles related to lipid metabolic, autophagic, and ferroptotic pathways were investigated. Palmitic acid (PA)-overload was used to assess lipotoxicity-induced necrosis. ResultsAs compared with controls, the Spike cells showed a significant increase in lipid depositions on cell membranes as well as dysregulation of expression of a panel of molecules involved lipid metabolism, autophagy, and ferroptosis. The Spike cells showed an upregulation of nuclear factor erythroid 2-related factor 2 (Nrf2), a multifunctional transcriptional factor, in response to PA. Furthermore, the Spike cells exhibited increased necrosis in response to PA-induced lipotoxicity compared to control cells in a time- and dose-dependent manner via ferroptosis, which could be attenuated by the Nrf2 inhibitor trigonelline. ConclusionsThe Spike protein impairs lipid metabolic and autophagic pathways in host cells, leading to increased susceptibility to lipotoxicity via ferroptosis which can be suppressed by a Nrf2 inhibitor. This data also suggests a central role of Nrf2 in Spike-induced lipid metabolic impairments. HighlightsO_LIThe Spike protein increases lipid deposition in host cell membranes C_LIO_LIThe Spike protein impairs lipid metabolic and autophagic pathways C_LIO_LIThe Spike protein exaggerates PA-induced lipotoxicity in host cells via ferroptosis C_LIO_LINrf2 inhibitor Trigonelline can mitigate the Spike protein-induced necrosis C_LI

cell biology↗

bioRxiv2022

Structure of IgG-Fc hexamer reveals a mutual lock-and-key mode of Fc-Fc interaction

IgG exists mainly as monomer, but recent studies suggest that IgG forms hexamer to mediate antibody functions. Although many structures of IgG monomer and its fragments are determined, there is no IgG hexamer structure at atomic level due to the weak Fc-Fc interactions. Here we engineered a hexameric IgG with IgM tailpiece fusion and determined the structure by cryo-EM. IgG-Fc hexamer forms hexagon symmetry with the six Fcs lie in a plane with Fc-Fc interaction in a mutual lock-and-key mode. This structure provides structural insights into Fc-Fc interaction of IgG and reveals molecular basis for its function.

biophysics↗

bioRxiv2022

Functional networks of the human bromodomain-containing proteins

BackgroundBromodomains are a structurally conserved epigenetic reader domain that bind to acetylated lysine residues in both histone and non-histone proteins. Bromodomain-containing proteins (BRD proteins) function are established scaffolds in the assembly of multi-protein complexes to regulate diverse biological processes. BRD proteins have been classified based on biological and functional similarity, however the functions of many BRD proteins remains unknown. PPI network analysis is useful for revealing organizational roles, identifying functional clusters, and predicting function for BRD proteins. ResultsWe used available data to construct protein-protein interaction networks (PPINs) to study the properties of the human bromodomain protein family. The network properties of the BRD PPIN establishes that the BRD proteins serve as hub proteins that are enriched near the global center to form an inter-connected PPIN. We identified dense subgraphs formed by BRD proteins and find that different BRD proteins share topological similarity and functional associations. We explored the functional relationships through clustering and Hallmark pathway gene set enrichment analysis and identify potential biological roles for different BRD proteins. ConclusionsIn our network analysis we confirmed that BRD proteins are conserved central nodes in the human PPI network and function as scaffolds to form distinctive functional clusters. Overall, this study provides detailed insight into the predictive functions of BRD proteins in the context of functional complexes and biological pathways.

bioinformatics↗

bioRxiv2022

cir-DNA sequencing revealed the landscape of extrachromosomal circular DNA in articular cartilage and the potential roles in osteoarthritis

ObjectiveExtrachromosomal circular DNA (eccDNA) has been shown to be involved in several physiological and pathological processes including immunity, inflammation, aging and tumor. However, the expression of eccDNA in cartilage has been not reported until now. In this study, we aimed to investigate the landscape of eccDNA in articular cartilage and analyze the potential roles in Osteoarthritis (OA). MethodsThe samples of articular cartilage were obtained from total knee arthroplasty (TKA) donors with OA. The mtDNAs (mitochondrial DNA) and the linear DNAs from chondrocytes of articular cartilage were removed. Then the eccDNAs were enriched for cir-DNA sequencing. After quality control evaluation, we systematically revealed the identified eccDNA data including size distribution, the size range and sequence pattern. Moreover, we explored and disscussed the potential roles of eccDNA in OA via motif analysis and GO/KEGG pathways analysis. ResultsThe chondrocytes from OA cartilage contained an abundance of eccDNAs, which was termed as OC-eccDNAs (OA Cartilage-derived eccDNA). The characteristics of OC-eccDNAs were tissue specific, including the distribution, the size range and sequence pattern. Moreover, the functional analysis indicated that eccDNA may be involved in the homeostasis maintenance of chondrocytes and participated in the process of OA. ConclutionsOur data firstly showed the landscape of eccDNA in articular cartilage, and preliminarily indicated the potential roles of eccDNA in OA.

molecular biology↗

bioRxiv2022

Tracing 3D apoplasmic complexity in vascular tissue of live conifer needles by X-ray computed tomography

Architecture and conducting area of vascular elements along conifer needles are fundamentally different from broad leaves. We hypothesised that the needles unique transfusion tissue offers different mechanisms for water allocation and used multimodal imaging to dissect the critical water exchange interfaces in this xerophytic leaf type. Our study examined intact conifer needles with X-ray computed tomography (XCT) and water-soluble tracers, allowing to render the functional 3D structure of the water-filled apoplast and the complementary symplasmic domain inside the bundle sheath. Segmentation of these data, together with fluorescence and electron microscopy of axial phloem and xylem elements along the needle, enabled quantification of the dimensions of the conducting tissue complex. The transfusion tracheid system between the endodermis-type bundle sheath and the axial venation formed a sponge-like apoplast domain. Transfusion parenchyma cell chains bridged this domain not directly but as tortuous symplasmic pathways between bundle sheath and axial phloem, which is nearly exclusively accessible at flanks. The transfusion tissue extends the plasma membrane surface for phloem loading and provides a large volume space. We discuss that this unique tissue plays an important role in the subtle interplay between water uptake/storage and sugar transport that has evolved to cope with desiccation stress.

plant biology↗

bioRxiv2021

Identification and characterization of BEND2 as a novel and key regulator of meiosis during mouse spermatogenesis

The chromatin state undergoes global and dynamic changes during spermatogenesis, and is critical to chromosomal synapsis, meiotic recombination, and transcriptional regulation. However, the key regulators involved and the underlying molecular mechanisms remain poorly understood. Herein we report that mouse BEND2, one of the BEN-domain-containing proteins conserved in vertebrates, was specifically expressed in spermatogenic cells within a short time-window spanning meiotic initiation, and that it plays an essential role in the progression of prophase in meiosis I. Bend2 gene knockout in male mice arrested meiosis at the transition from zygonema to pachynema, disrupted synapsis and DNA double-strand break repair, and induced non-homologous chromosomal pairing. BEND2 interacted with a number of chromatin-associated proteins--including ZMYM2, LSD1, CHD4, and ADNP-- which are components of certain transcription-repressor complexes. BEND2-binding sites were identified in diverse chromatin states and enriched in simple sequence repeats. BEND2 contributed to shutting down the mitotic gene-expression program and to the activation of meiotic and post-meiotic gene expression, and it regulated chromatin accessibility as well as the modification of H3K4me3. Therefore, our study identified BEND2 as a novel and key regulator of meiosis, gene expression, and chromatin state during mouse spermatogenesis. TeaserMeiosis is a highly complex yet poorly understood process that involves the concerted actions of an increasing number of regulators, of which the list remains incomplete. Ma et al. identified BEND2 as a novel and key regulator of meiosis and showed that it interacts with critical chromatin modulators and specific genomic elements to control the expression of mitotic and meiotic genes.

developmental biology↗

bioRxiv2021

The crystal structure of vaccinia virus protein E2 and perspectives on the prediction of novel viral protein folds

The morphogenesis of vaccinia virus (VACV, family Poxviridae), the smallpox vaccine, is a complex process involving multiple distinct cellular membranes and resulting in multiple different forms of infectious virion. Efficient release of enveloped virions, which promote systemic spread of infection within hosts, requires the VACV protein E2 but the molecular basis of E2 function remains unclear and E2 lacks sequence homology to any well-characterised family of proteins. We solved the crystal structure of VACV E2 to 2.3 [A] resolution, revealing that it comprises two domains with novel folds: an N-terminal annular (ring) domain and a C-terminal head domain. The C-terminal head domain displays weak structural homology with cellular (pseudo)kinases but lacks conserved surface residues or kinase features, suggesting that it is not enzymatically active, and possesses a large surface basic patch that might interact with phosphoinositide lipid headgroups. Recent deep learning methods have revolutionised our ability to predict the three-dimensional structures of proteins from primary sequence alone. VACV E2 is an exemplar difficult viral protein target for structure prediction, being comprised of multiple novel domains and lacking sequence homologues outside Poxviridae. AlphaFold2 nonetheless succeeds in predicting the structures of the head and ring domains with high and moderate accuracy, respectively, allowing accurate inference of multiple structural properties. The advent of highly accurate virus structure prediction marks a step-change in structural virology and beckons a new era of structurally-informed molecular virology.

microbiology↗

bioRxiv2021

miR2105 regulates ABA biosynthesis via OsbZIP86-OsNCED3 module to contribute to drought tolerance in rice

Induced abscisic acid (ABA) biosynthesis plays an important role in plant tolerance to abiotic stresses, including drought, cold and salinity. However, regulation pathway of the ABA biosynthesis in response to stresses is unclear. Here, we identified a rice miRNA, osa-miR2105 (miR2105), which plays a crucial role in ABA biosynthesis under drought stress. Analysis of expression, transgenic rice and cleavage site showed that OsbZIP86 is a target gene of miR2105. Subcellular localization and luciferase activity assays showed that OsbZIP86 is a nuclear transcription factor. In vivo and in vitro analyses showed that OsbZIP86 directly binds to the promoter of OsNCED3, and interacts with OsSAPK10, resulting in enhanced-expression of OsNCED3. Transgenic rice plants with knock-down of miR2105 or overexpression of OsbZIP86 showed higher ABA content, more tolerance to drought, a lower rate of water loss, more stomatal closure than wild type rice ZH11 under drought stress. These rice plants showed no penalty with respect to agronomic traits under normal conditions. By contrast, transgenic rice plants with miR2105 overexpression, OsbZIP86 downregulation, or OsbZIP86 knockout displayed less tolerance to drought stress and other phenotypes. Collectively, our results show that a regulatory network of miR2105-OsSAPK10/OsbZIP86-OsNCED3 control ABA biosynthesis in response to drought stress. One-sentence summary miR2105-OsbZIP86-OsNCED3 module plays crucial role in mediating ABA biosynthesis to contribute to drought tolerance with no penalty with respect to agronomic traits under normal conditions.

molecular biology↗

bioRxiv2021

Concomitant Pyroptotic and Apoptotic Cell Death Triggered in Monocytes Infected by Zika Virus

Zika virus (ZIKV) is a positive-sense RNA flavivirus and can cause serious neurological disorders including microcephaly in infected fetus. As a mosquito-borne arbovirus, ZIKV enters bloodstream and is transmitted into the fetus through the placenta in pregnant women. Monocytes are considered one of the earliest blood cell types to be infected by ZIKV. As a first line defence, monocytes are crucial components in innate immunity and host responses and may impact viral pathogenesis in humans. Previous studies have shown that ZIKV infection can activate inflammasomes and induce proinflammatory cytokines in monocytes. In this report, we showed that ZIKV carried out a productive infection, which lead to cell death in human and murine monocytic cells. In addition to the presence of cleaved caspase-3, indicating that apoptosis was involved, we identified the cleaved caspase-1 and gasdemin D (GSDMD) as well as increased secretion of IL-1{beta} and IL-18, suggesting that the inflammasome was activated that may lead to pyroptosis in infected monocytes. The pyroptosis was NLRP3-dependent and could be suppressed in the monocytes treated with shRNA to target and knockdown caspase-1, or an inhibited for caspase-1, indicating that the pyroptosis was triggered via a canonical approach. Our findings in this study demonstrate a concomitant occurrence of apoptosis and pyroptosis in ZIKV-infected monocytes, with multiple mechanisms involved in the cell death, which may have potentially significant impacts on viral pathogenesis in humans.

microbiology↗

bioRxiv2021

Global, Low Amplitude Cortical State Predicts Response Outcomes in a Selective Detection Task

Spontaneous neuronal activity strongly impacts stimulus encoding and behavioral responses. We sought to determine the effects of neocortical prestimulus activity on stimulus detection. We trained mice in a selective whisker detection task, in which they learned to respond (lick) to target stimuli in one whisker field and ignore distractor stimuli in the contralateral whisker field. During expert task performance, we used widefield Ca2+ imaging to assess prestimulus and post-stimulus neuronal activity broadly across frontal and parietal cortices. We found that lower prestimulus activity correlated with enhanced stimulus detection: lower prestimulus activity predicted response versus no response outcomes and faster reaction times. The activity predictive of trial outcome was distributed through dorsal neocortex, rather than being restricted to whisker or licking regions. Using principal component analysis, we demonstrate that response trials are associated with a distinct and less variable prestimulus neuronal subspace. For single units, prestimulus choice probability was weak yet distributed broadly, with lower than chance choice probability correlating with stronger sensory and motor encoding. These findings support a low amplitude, low variability, optimal prestimulus cortical state for stimulus detection that presents globally and predicts response outcomes for both target and distractor stimuli.

neuroscience↗

bioRxiv2021

Lifestyle and Genetic Factors Modify Parent-of-Origin Effects on the Human Methylome

Backgroundparent-of-origin effects (POE) play important roles in development and complex disease and thus understanding their regulation and associated molecular and phenotypic variation are warranted. Previous studies have mainly focused on the detection of genomic regions or phenotypes regulated by POE. Understanding whether POE may be modified by environmental or genetic exposures is important for understanding of the source of POE-associated variation, but only a few case studies addressing these modifiable POE exist. Methodsin order to understand this high order of POE regulation, we screened 101 genetic and environmental factors such as "predicted mRNA expression levels" of DNA methylation/imprinting machinery genes and early/late lifestyle/environmental exposures. POE-mQTL-modifier interaction models were proposed to test the potential of these factors to modify POE at DNA methylation using data from Generation Scotland: The Scottish Family Health Study(N=2315). Resultsa set of vulnerable/modifiable POE-CpGs were identified (modifiable-POE-regulated CpGs, N=3). Four factors, "lifetime smoking status" and "predicted mRNA expression levels" of TET2, SIRT1 and KDM1A, were found to significantly modify the POE on the three CpGs in both discovery and replication datasets. Importantly, the POE on one of the CpGs were modified by both genetic and environmental factors. We further identified plasma protein and health-related phenotypes associated with the methylation level of one of the identified CpGs. Conclusionsthe modifiable POE identified here revealed an important yet indirect path through which genetic background and environmental exposures introduce their effect on DNA methylation, motivating future comprehensive evaluation of the role of these modifiers in complex diseases. Key MessagesO_LIPrevious population studies showed that parent-origin-effects(POE) on human methylome can be widespread and affect health-related traits and diseases. C_LIO_LIWhether the POE remained stable throughout the life or can be modified by genetic or environmental factors were largely unknown. C_LIO_LIBy systematically screening 101 genetic and environmental factors in a large cohort(GS:SFHS) we provided the first population-level replicated evidence that those measuring lifestyle (smoking) and predicted expression of DNA methylation- or imprinting-machinery genes are amongst the factors that can modulate the POE of mQTLs for a set of CpG sites. C_LIO_LIWe found those modifiable-POE-regulated CpGs are also phenotypically relevant - one is associated with the plasma levels of CLEC4C and health-related phenotypes such as HDL levels. C_LIO_LIThe modifiable POE identified here revealed an important yet indirect path through which genetic background and environmental exposures introduce their effect on DNA methylation and their potential phenotypic consequences. This also provided a paradigm for further studies to explore how environmental and genetic effects can be integrated at methylation level. C_LI

genetics↗

bioRxiv2021

MicroRNA-202 prevents precocious spermatogonial differentiation and meiotic initiation during mouse spermatogenesis

Spermatogonial differentiation and meiotic initiation during spermatogenesis are tightly regulated by a number of genes including those coding enzymes for miRNA biogenesis. However, whether and how single miRNAs regulate these processes remain unclear. Here, we report that miR-202, a member of the let-7 family, prevents precocious spermatogonial differentiation and meiotic initiation in spermatogenesis by regulating the timely expression of many genes including those for other key regulators. In miR-202 knockout (KO) mice, the undifferentiated spermatogonial pool is reduced, ultimately causing agametic seminiferous tubules. SYCP3, STRA8 and DMRT6 are expressed earlier in KO mice than in wild-type (WT) littermates, and Dmrt6 mRNA is a direct target of miR-202-5p. Moreover, the precocious spermatogonial differentiation and meiotic initiation were also observed in KO spermatogonial stem cells when cultured and induced in vitro, and could be rescued by the knockdown of Dmrt6. Therefore, we have not only shown that miR-202 is a novel regulator of meiotic initiation but also added a new module to the underlying regulatory network. Summary statementA single miRNA, miR-202, prevents precocious differentiation and meiotic initiation during spermatogenesis. miR-202, DMRT6 and STRA8 act together as a novel module in the regulatory network of meiotic initiation.

developmental biology↗

bioRxiv2021

Uncovering the Genetic Profiles Underlying the Intrinsic Organization of the Human Cerebellum

The functional diversity of the human cerebellum is largely believed to be derived more from its extensive connections rather than being limited to its mostly invariant architecture. However, whether and how the determination of cerebellar connections in its intrinsic organization interact with microscale gene expression is still unknown. Here we decode the genetic profiles of the cerebellar functional organization by investigating the genetic substrates simultaneously linking cerebellar functional heterogeneity and its drivers, i.e., the connections. We not only identified 443 network-specific genes but also discovered that their co-expression pattern correlated strongly with intra-cerebellar functional connectivity (FC). Ninety of these genes were also linked to the FC of cortico-cerebellar cognitive-limbic networks. To further discover the biological functions of these genes, we performed a "virtual gene knock-out" by observing the change in the coupling between gene co-expression and FC and divided the genes into two subsets, i.e., a positive gene contribution indicator (GCI+) involved in cerebellar neurodevelopment and a negative gene set (GCI-) related to neurotransmission. A more interesting finding is that GCI- is significantly linked with the cerebellar connectivity-behavior association and many recognized brain diseases that are closely linked with the cerebellar functional abnormalities. Our results could collectively help to rethink the genetic substrates underlying the cerebellar functional organization and offer possible micro-macro interacted mechanistic interpretations of the cerebellum-involved high order functions and dysfunctions in neuropsychiatric disorders.

neuroscience↗

bioRxiv2021

Engineered AIM-based Selective Autophagy to Degrade Proteins and Organelles

Techniques for disrupting of protein function are essential for biological researches and therapeutics development. Though well-established, genetic perturbation strategies may have off-target effects and/or trigger compensatory mechanisms, and cannot efficiently eliminate existing protein variants or aggregates1, 2. Therefore, precise and direct protein-targeting methods are highly desired. Here we describe a novel method for targeted protein clearance by engineering an autophagy receptor with a binder to provide target specificity and an ATG8-binding motif (AIM) to link the targets to nascent autophagosomes, thus harnessing the autophagy machinery for degradation. We demonstrate its specificity and broad potentials by degrading various fluorescent-tagged proteins and peroxisome organelle, using a tobacco-based transient expression system, and by degrading endogenous proteins in transgenic Arabidopsis expressing engineered receptors. With the wide substrate scope and specificity of selective autophagy, our method provides a convenient and robust strategy for eliminating proteins and aggregates, and may enable developing new treatments for protein-related disorders.

cell biology↗

bioRxiv2021

Variable Induction of Pro-inflammatory Cytokines by Commercial SARS CoV-2 Spike Protein Reagents: Potential Impacts of LPS on In Vitro Modeling and Pathogenic Mechanisms In Vivo

Proinflammatory cytokine production following infection with severe acute respiratory syndrome coronavirus 2 (SARS CoV-2) is associated with poor clinical outcomes. Like SARS CoV-1, SARS CoV-2 enters host cells via its spike protein, which attaches to angiotensin-converting enzyme 2 (ACE2). As SARS CoV-1 spike protein is reported to induce cytokine production, we hypothesized that this pathway could be a shared mechanism underlying pathogenic immune responses. We herein compared the capabilities of Middle East Respiratory Syndrome (MERS), SARS CoV-1 and SARS CoV-2 spike proteins to induce cytokine expression in human peripheral blood mononuclear cells (PBMC). We observed that only specific commercial lots of SARS CoV-2 induce cytokine production. Surprisingly, recombinant SARS CoV-2 spike proteins from different vendors and batches exhibited different patterns of cytokine induction, and these activities were not inhibited by blockade of spike protein-ACE2 binding using either soluble ACE2 or neutralizing anti-S1 antibody. Moreover, commercial spike protein reagents contained varying levels of endotoxin, which correlated directly with their abilities to induce cytokine production. The lipopolysaccharide (LPS) inhibitor, polymyxin B, blocked this cytokine induction activity. In addition, SARS CoV-2 spike protein avidly bound soluble LPS in vitro, rendering it a cytokine inducer. These results not only suggest caution in monitoring the purity of SARS CoV-2 spike protein reagents, but they indicate the possibility that interactions of SARS CoV-2 spike protein with LPS from commensal bacteria in virally infected mucosal tissues could promote pathogenic inflammatory cytokine production.

microbiology↗

bioRxiv2021

Flexibility-aware graph-based algorithm improves antigen epitopes identification

Epitopes of an antigen are the surface residues in the spatial proximity that can be recognized by antibodies. Identifying such residues has shown promising potentiality in vaccine design, drug development and chemotherapy, thus attracting extensive endeavors. Although great efforts have been made, the epitope prediction performance is still unsatisfactory. One possible issue accounting to this poor performance could be the ignorance of structural flexibility of antigens. Flexibility is a natural characteristic of antigens, which has been widely reported. However, this property has never been used by existing models. To this end, we propose a novel flexibility-aware graph-based computational model to identify epitopes. Unlike existing graph-based approaches that take the static structures of antigens as input, we consider all possible variations of the side chains in graph construction. These flexibility-aware graphs, of which the edges are highly enriched, are further partitioned into subgraphs by using a graph clustering algorithm. These clusters are subsequently expanded into larger graphs for detecting overlapping residues between epitopes if exist. Finally, the expanded graphs are classified as epitopes or non-epitopes via a newly designed graph convolutional network. Experimental results show that our flexibility-aware model markedly outperforms existing approaches and promotes the F1-score to 0.656. Comparing to the state-of-the-art, our approach makes an increment of F1-score by 16.3%. Further in-depth analysis demonstrates that the flexibility-aware strategy contributes the most to the improvement. The source codes of the proposed model is freely available at https://github.com/lzhlab/epitope. Author summaryEpitope prediction is helpful to many biomedical applications so that dozens of models have been proposed aiming at improving prediction efficiency and accuracy. However, the performances are still unsatisfactory due to its complicated nature, particularly the noteworthy flexible structures, which makes the precise prediction even more challenging. The existing approaches have overlooked the flexibility during model construction. To this end, we propose a graph model with flexibility heavily involved. Our model is mainly composed of three parts: i) flexibility-aware graph construction; ii) overlapping subgraph clustering; iii) graph convolutional network-based subgraph classification. Experimental results show that our newly proposed model markedly outperforms the existing best ones, making an increment of F1-score by 16.3%.

bioinformatics↗