bioRxiv Science⌕ Search

Biology subjects

Qi, Y.

Publications and source records attributed to Qi, Y..

At least 163 records · Page 9Linked to original sources
bioRxiv2020

Transfer Learning with MotifTransformers for Predicting Protein-Protein Interactions Between a Novel Virus and Humans

Viruses such as SARS-CoV-2 infect the human body by forming interactions between virus proteins and human proteins. However, experimental methods to find protein interactions are inadequate: large scale experiments are noisy, and small scale experiments are slow and expensive. Inspired by the recent successes of deep neural networks, we hypothesize that deep learning methods are well-positioned to aid and augment biological experiments, hoping to help identify more accurate virus-host protein interaction maps. Moreover, computational methods can quickly adapt to predict how virus mutations change protein interactions with the host proteins. We propose DeepVHPPI, a novel deep learning framework combining a self-attention-based transformer architecture and a transfer learning training strategy to predict interactions between human proteins and virus proteins that have novel sequence patterns. We show that our approach outperforms the state-of-the-art methods significantly in predicting Virus-Human protein interactions for SARS-CoV-2, H1N1, and Ebola. In addition, we demonstrate how our framework can be used to predict and interpret the interactions of mutated SARS-CoV-2 Spike protein sequences. AvailabilityWe make all of our data and code available on GitHub https://github.com/QData/DeepVHPPI. ACM Reference FormatJack Lanchantin, Tom Weingarten, Arshdeep Sekhon, Clint Miller, and Yanjun Qi. 2021. Transfer Learning for Predicting Virus-Host Protein Interactions for Novel Virus Sequences. In Proceedings of ACM Conference (ACM-BCB). ACM, New York, NY, USA, 10 pages. https://doi.org/??

bioinformatics↗
bioRxiv2020

A time-course study of actively stained mouse brains: DTI parameter and connectomic stability over one year

While the application of diffusion tensor imaging (DTI), tractography, and connectomics to fixed ex-vivo tissue is a common practice today, there have been limited studies examining the effects of fixation on brain microstructure over extended periods. This time-course study reports the changes of regional brain volumes and diffusion scalar parameters, such as fractional anisotropy across twelve representative brain regions as measures of brain structural stability. The scalar DTI parameters and regional volumes were highly variable over the first two weeks after fixation. The same parameters were stable over a two to eight-week window after fixation which means confounds from tissue stability over that scanning window are minimal. Quantitative connectomes were analyzed over the same time period with extension out to one year. While there is some change in the scalar metrics at one year after fixation, these changes are sufficiently small, particularly in white matter to support reproducible connectomes over a period ranging from two weeks to one year post fixation. These findings delineate a stable scanning period during which brain volumes, diffusion scalar metrics and connectomes are remarkably stable.

neuroscience↗
bioRxiv2020

Characterization and structural basis of a lethal mouse-adapted SARS-CoV-2

The ongoing SARS-CoV-2 pandemic has brought an urgent need for animal models to study the pathogenicity of the virus. Herein, we generated and characterized a novel mouse-adapted SARS-CoV-2 strain, named MASCp36, that causes severe acute respiratory symptoms and mortality in standard laboratory mice. Particularly, this model exhibits age and gender related skewed distribution of mortality akin to severe COVID-19, and the 50% lethal dose (LD50) of MASCp36 was 58 PFU in 9-month-old, male BALB/c mice. Deep sequencing identified three amino acid substitutions, N501Y, Q493H, and K417N, subsequently emerged at the receptor binding domain (RBD) of MASCp36, during in vivo passaging. All three mutations in RBD significantly enhanced the binding affinity to its endogenous receptor, mouse ACE2 (mACE2). Cryo-electron microscopy (cryo-EM) analysis of human ACE2 (hACE2) or mACE2 in complex with the RBD of MASCp36 at 3.1 to 3.7 angstrom resolution elucidates molecular basis for the receptor-binding switch driven by specific amino acid substitutions. Interestingly, N501Y and Q493H enhanced the binding affinity to human ACE2 (hACE2); while triple mutations N501Y/Q493H/K417N decreased affinity to hACE2, thus led to the reduced infectivity of MASCp36 to human cells. Our study not only provides a robust platform for studying the pathogenesis of severe COVID-19 and rapid evaluation of coutermeasures against SARS-CoV-2, but also unveils the molecular mechanism for the rapid adaption and evolution of SARS-CoV-2 in human and animals. One sentence summaryA mouse adapted SARS-CoV-2 strain that harbored specific amino acid substitutions in the RBD of S protein showed 100% mortality in aged, male BALB/c mice.

microbiology↗
bioRxiv2020

Flexible and transparent silver nanowire structures for multifunctional electrical and optical biointerfacing

Transparent microelectrodes have recently emerged as a promising approach to combine electrophysiology with optophysiology for multifunctional biointerfacing. High-performance flexible platforms that allow seamless integration with soft tissue systems for such applications are urgently needed. Here, silver nanowires (Ag NWs)-based transparent microelectrodes and interconnects are designed to meet this demand. The Ag NWs percolating networks are patterned on flexible polymer substrates using an innovative photolithography-based solution-processing technique. The resulting nanowire networks exhibit a high average optical transparency of 76.1-90.0% over the visible spectrum, low normalized electrochemical impedance of 3.4-15 {Omega} cm2 at 1 kHz which is even better than those of opaque solid Ag films, superior sheet resistance of 11-25 {Omega} sq-1, excellent mechanical stability up to 10,000 bending cycles, good biocompatibility and chemical stability. Studies on Langendorff-perfused mouse and rat hearts demonstrate that the Ag NWs microelectrodes enable high-fidelity real-time monitoring of heart rhythm during co-localized optogenetic pacing and optical mapping with negligible light-induced electrical artifacts. This proof-of-concept work illustrates that the solution-processed, transparent, and flexible Ag NWs networks are a promising candidate for the next-generation of large-area multifunctional biointerfaces for interrogating complex biological systems in basic and translational research.

bioengineering↗
bioRxiv2020

Systematic discovery and functional interrogation of SARS-CoV-2 viral RNA-host protein interactions during infection

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the cause of a pandemic with growing global mortality. There is an urgent need to understand the molecular pathways required for host infection and anti-viral immunity. Using comprehensive identification of RNA-binding proteins by mass spectrometry (ChIRP-MS), we identified 309 host proteins that bind the SARS-CoV-2 RNA during active infection. Integration of this data with viral ChIRP-MS data from three other positive-sense RNA viruses defined pan-viral and SARS-CoV-2-specific host interactions. Functional interrogation of these factors with a genome-wide CRISPR screen revealed that the vast majority of viral RNA-binding proteins protect the host from virus-induced cell death, and we identified known and novel anti-viral proteins that regulate SARS-CoV-2 pathogenicity. Finally, our RNA-centric approach demonstrated a physical connection between SARS-CoV-2 RNA and host mitochondria, which we validated with functional and electron microscopy data, providing new insights into a more general virus-specific protein logic for mitochondrial interactions. Altogether, these data provide a comprehensive catalogue of SARS-CoV-2 RNA-host protein interactions, which may inform future studies to understand the mechanisms of viral pathogenesis, as well as nominate host pathways that could be targeted for therapeutic benefit. Highlights{middle dot} ChIRP-MS of SARS-CoV-2 RNA identifies a comprehensive viral RNA-host protein interaction network during infection across two species {middle dot} Comparison to RNA-protein interaction networks with Zika virus, dengue virus, and rhinovirus identify SARS-CoV-2-specific and pan-viral RNA protein complexes and highlights distinct intracellular trafficking pathways {middle dot} Intersection of ChIRP-MS and genome-wide CRISPR screens identify novel SARS-CoV-2-binding proteins with pro- and anti-viral function {middle dot} Viral RNA-RNA and RNA-protein interactions reveal specific SARS-CoV-2-mediated mitochondrial dysfunction during infection

microbiology↗
bioRxiv2020

In-depth analysis reveals complex molecular etiology of idiopathic cerebral palsy

Cerebral palsy (CP), the most prevalent physical disability in children, has long been ignored with regard to its inherent molecular mechanism. In this work, we performed in-depth clinical and molecular analysis on 120 idiopathic CP families, and identified in half of the patients the underlying risk factors. By a compilation of 117 CP-related genes, we recognized the characteristic features in terms of inheritance and function, and proposed a dichotomous classification system according to the expression patterns. In two patients with both CP and intellectual disability, we revealed that the defective TYW1, a tRNA hypermodification enzyme, caused microcephaly and problems in motion and cognition by hindering neuronal proliferation and migration. We developed an algorithm and proved in mice brain that this malfunctioning hypermodification specifically perturbed the translation of a subset of proteins involved in cell cycling. In a CP patient with normal intelligence, we identified a mitochondrial enzyme GPAM, the hypomorphic form of which led to hypomyelination of corticospinal tract. We confirmed that the aberrant Gpam in mice perturbed the lipid metabolism in astrocyte, resulting in suppressed astrocyte proliferation and a shortage of lipid contents supplied for oligodendrocyte myelination. This work broadened the scope of understanding of CP etiology and paved a way for innovative therapeutic strategies.

genetics↗
bioRxiv2020

Global proteome and phosphoproteome alterations in third generation EGFR TKI resistance reveal drug targets to circumvent resistance

Lung cancer is the leading cause of cancer mortality worldwide. The treatment of lung cancer patients harboring mutant EGFR with orally administered EGFR TKIs has been a paradigm shift. Osimertinib and rociletinib are 3rd generation irreversible EGFR TKIs targeting the EGFR T790M mutation. Osimertinib is the current standard care for patients with EGFR mutations due to increased efficacy, lower side effects, and enhanced brain penetrance. Unfortunately, all patients develop resistance. Genomic approaches have primarily been used to interrogate resistance mechanisms. Here, we have characterized the proteome and phosphoproteome of a series of isogenic EGFR mutant lung adenocarcinoma cell lines that are either sensitive or resistant to these drugs. To our knowledge, this is the most comprehensive proteomic dataset resource to date to investigate 3rd generation EGFR TKI resistance in lung adenocarcinoma. We have interrogated this unbiased global quantitative mass spectrometry dataset to uncover alterations in signaling pathways, reveal a proteomic signature of epithelial mesenchymal transition (EMT) and identify kinases and phosphatases with altered expression and phosphorylation in TKI resistant cells. Decreased tyrosine phosphorylation of key sites in the phosphatase SHP2 suggests its inhibition resulting in inhibition of RAS/MAPK and activation of PI3K/AKT pathways. Furthermore, we performed anticorrelation analyses of this phosphoproteomic dataset with the published drug-induced P100 phosphoproteomic datasets from the Library of Integrated Network-Based Cellular Signatures (LINCS) program to predict drugs with the potential to overcome EGFR TKI resistance. We identified dactolisib, a PI3K/mTOR inhibitor, which in combination with osimertinib overcomes resistance both in vitro and in vivo.One Sentence Summary Global quantitative proteome and phosphoproteome analyses to examine altered signaling pathways in isogenic 3rd generation EGFR TKI sensitive and resistant cells.Competing Interest StatementThe authors have declared no competing interest.View Full Text

cancer biology↗
bioRxiv2020

Hippocampal hyperactivity in a rat model of Alzheimers disease

Neuronal network dysfunction is a hallmark of Alzheimers disease (AD). However, the underlying pathomechanisms remain unknown. We analyzed the hippocampal micronetwork in a rat model of AD at an early disease stage at the beginning of extracellular amyloid beta (A{beta}) deposition. We established two-photon Ca2+-imaging in vivo in the hippocampus of rats and found hyperactivity of CA1 neurons. Patch-clamp recordings in brain slices in vitro revealed changes in the passive properties and intrinsic excitability of CA1 pyramidal neurons. Furthermore, we observed increased neuronal input resistance and prolonged action potential width in CA1 pyramidal neurons. Surprisingly, all parameters measured to quantify synaptic inhibition and excitation onto CA1 pyramidal neurons were intact suggesting a cell immanent deficit. Our data support the view that altered intrinsic excitability of CA1 neurons may precede inhibitory dysfunction at an early stage of disease progression.

neuroscience↗
bioRxiv2020

Developmentally Delayed Epigenetic Reprogramming Underlying the Pathogenesis of Preeclampsia

Preeclampsia, a life-threatening pregnancy complication characterized by hypertension and multiorgan damage, affects 2-5% of pregnancies and causes 76,000 deaths per year. Most preeclampsia associated syndromes immediately dispel after removal of placenta, indicating a casual role of placenta in the pathogenesis. Failed transformation of spiral artery due to insufficient invasion and excessive apoptosis of trophoblast suggested developmental defects in preeclampsia placenta. However, the underlying molecular mechanisms that affected placenta development in preeclampsia remained elusive. Here we show that, in preeclampsia placenta, the epigenetic landscape formed during extraembryonic tissue differentiation was disrupted: dramatic chromatin accessibility shift affected known and novel genes implicated in preeclampsia. DNA methylation defects in preeclampsia affected lineage-defining PcG-controlled loci in trophectoderm. LTR12 retrotransposons associated with VCT/SCT-specific genes were hypermethylated. Meanwhile, hundreds of PcG-regulated EVT-specific gene promoters, which otherwise undergone post-ZGA extraembryonic-tissue-specific de novo methylation, were hypomethylated and hyper-activated. Together, these epigenetic defects resulted in placenta developmental delay in preeclampsia. The defective methylation pattern could be detected in serum cfDNA, and could be used to accurately predict preeclampsia at early pregnancy weeks in independent validation cohorts. Our data suggests that the preeclampsia placenta represents a stalled state of epigenetic reprogramming en route of development from trophectoderm to normal placenta.

genomics↗
bioRxiv2020

FastSK: Fast Sequence Analysis with Gapped String Kernels

Gapped k-mer kernels with Support Vector Machines (gkm-SVMs) have achieved strong predictive performance on regulatory DNA sequences on modestly-sized training sets. However, existing gkm-SVM algorithms suffer from slow kernel computation time, as they depend exponentially on the sub-sequence feature-length, number of mismatch positions, and the tasks alphabet size. In this work, we introduce a fast and scalable algorithm for calculating gapped k-mer string kernels. Our method, named FastSK, uses a simplified kernel formulation that decomposes the kernel calculation into a set of independent counting operations over the possible mismatch positions. This simplified decomposition allows us to devise a fast Monte Carlo approximation that rapidly converges. FastSK can scale to much greater feature lengths, allows us to consider more mismatches, and is performant on a variety of sequence analysis tasks. On 10 DNA transcription factor binding site (TFBS) prediction datasets, FastSK consistently matches or outperforms the state-of-the-art gkmSVM-2.0 algorithms in AUC, while achieving average speedups in kernel computation of [~]100x and speedups of [~]800x for large feature lengths. We further show that FastSK outperforms character-level recurrent and convolutional neural networks across all 10 TFBS tasks. We then extend FastSK to 7 English-language medical named entity recognition datasets and 10 protein remote homology detection datasets. FastSK consistently matches or outperforms these baselines. Our algorithm is available as a Python package and as C++ source code1.

bioinformatics↗
bioRxiv2020

The beginning of the end: a chromosomal assembly of the New World malaria mosquito ends with a novel telomere

Chromosome level assemblies are accumulating in various taxonomic groups including mosquitoes. However, even in the few reference-quality mosquito assemblies, a significant portion of the heterochromatic regions including telomeres remain unresolved. Here we produce a de novo assembly of the New World malaria mosquito, Anopheles albimanus by integrating Oxford Nanopore sequencing, Illumina, Hi-C and optical mapping. This 172.6 Mbps female assembly, which we call AalbS3, is obtained by scaffolding polished large contigs (contig N50=13.7 Mbps) into three chromosomes. All chromosome arms end with telomeric repeats, which is the first in mosquito assemblies and represents a significant step towards the completion of a genome assembly. These telomeres consist of tandem repeats of a novel 30-32 bp telomeric repeat unit (TRU) and are confirmed by analysing the termini of long reads and through both chromosomal in situ hybridization and a Bal31 sensitivity assay. The AalbS3 assembly included previously uncharacterized centromeric and rDNA clusters and more than doubled the content of transposable elements and other repetitive sequences. This telomere-to-telomere assembly, although still containing gaps, represents a significant step towards resolving biologically important but previously hidden genomic components. The comparison of different scaffolding methods will also inform future efforts to obtain reference-quality genomes for other mosquito species. 100-word Article SummaryWe report AalbS3, a telomere-to-telomere assembly of the Anopheles albimanus genome produced by integrating advancing technologies including Oxford Nanopore and Bionano optical mapping. AalbS3 features much of the difficult-to-assemble genomic dark matters including previously missed transposons, centromeres and rDNA clusters. We describe novel telomeric repeats that are confirmed by analysis of long reads and by telomere hybridization assays. This reference-quality assembly represents a significant step towards completing the genomic puzzle pieces and informs efforts to improve the assembly of other mosquito species. Future research into the relationship between telomere and mosquito life span may have significant implications to disease control.

genomics↗
bioRxiv2020

Data-driven polymer model for mechanistic exploration of diploid genome organization

Chromosomes are positioned non-randomly inside the nucleus to coordinate with their transcriptional activity. The molecular mechanisms that dictate the global genome organization and the nuclear localization of individual chromosomes are not fully understood. We introduce a polymer model to study the organization of the diploid human genome: it is data-driven as all parameters can be derived from Hi-C data; it is also a mechanistic model since the energy function is explicitly written out based on a few biologically motivated hypotheses. These two features distinguish the model from existing approaches and make it useful both for reconstructing genome structures and for exploring the principles of genome organization. We carried out extensive validations to show that simulated genome structures reproduce a wide variety of experimental measurements, including chromosome radial positions and spatial distances between homologous pairs. Detailed mechanistic investigations support the importance of both specific inter-chromosomal interactions and centromere clustering for chromosome positioning. We anticipate the polymer model, when combined with Hi-C experiments, to be a powerful tool for investigating large scale rearrangements in genome structure upon cell differentiation and tumor progression.

biophysics↗
bioRxiv2020

Aberrant pathogenic GM-CSF+ T cells and inflammatory CD14+CD16+ monocytes in severe pulmonary syndrome patients of a new coronavirus

Pathogenic human coronavirus infections, such as severe acute respiratory syndrome CoV (SARS-CoV) and Middle East respiratory syndrome CoV (MERS-CoV), cause high morbidity and mortality 1,2. Recently, a severe pneumonia-associated respiratory syndrome caused by a new coronavirus was reported at December 2019 (2019-nCoV) in the city Wuhan, Hubei province, China3-5, which was also named as pneumonia-associated respiratory syndrome (PARS)6. Up to 9th of February 2020, at least 37, 251 cases have been reported with 812 fatal cases according to the report from China CDC. However, the immune mechanism that potential orchestrated acute mortality from patients of 2019-nCoV is still unknown. Here we show that after the 2019-nCoV infection, CD4+T lymphocytes are rapidly activated to become pathogenic T helper (Th) 1 cells and generate GM-CSF etc. The cytokines environment induces inflammatory CD14+CD16+ monocytes with high expression of IL-6 and accelerates the inflammation. These aberrant and excessive immune cells may enter the pulmonary circulation in huge numbers and play an immune damaging role to causing lung functional disability and quick mortality. Our results demonstrate that excessive non-effective host immune responses by pathogenic T cells and inflammatory monocytes may associate with severe lung pathology. Therefore, we suggest that monoclonal antibody that targets the GM-CSF or interleukin 6 receptor may potentially curb immunopathology caused by 2019-nCoV and consequently win more time for virus clearance.

immunology↗
bioRxiv2020

A novel function of sphingosine kinase 2 in the metabolism of sphinga-4,14-diene lipids.

The number, position, and configuration of double bonds in lipid acyl chains affects membrane packing, fluidity, and recruitment of signalling proteins. Studies on mammalian sphingolipids have focused on those with a saturated sphinganine or mono-unsaturated sphingosine long chain base. Sphingolipids with a diunsaturated sphingadiene base have been reported but are poorly characterised. Employing high-resolution untargeted mass spectrometry, we observed marked accumulation of lipids containing a sphingadiene base, but not those with a more common sphingosine backbone, in the hippocampus of mice lacking the metabolic enzyme sphingosine kinase 2 (SphK2). Applying ultraviolet photodissociation tandem mass spectrometry (UVPD-MS/MS) the double bonds were confidently assigned to the C4-C5 and C14-C15 positions of the sphingoid base. Sphingosine kinases are involved in lysosomal catabolism of all sphingolipids, producing sphingoid base phosphates that are irreversibly degraded by sphingosine 1-phosphate lyase. Both SphK1 and SphK2 phosphorylated sphinga-4,14-diene as efficiently as sphingosine, however deuterated tracer experiments demonstrated that ceramides with a sphingosine base are more rapidly metabolised in cultured cells than those with a sphingadiene base. SphK2 silencing significantly impeded the catabolism of both sphingosine- and sphingadiene-based sphingolipids. Since SphK2 is the dominant sphingosine kinase in brain, we propose that accumulation of sphingadiene lipids in SphK2-deficient brains results from the intrinsically slower catabolism of sphingadiene lipids, combined with a bottleneck in the catabolic pathway created by the absence of SphK2. We speculate that accumulation of these lipids in the absence of SphK2 function may affect the fluidity and signalling properties of cell membranes.

biochemistry↗
bioRxiv2020

A Generalized Discrete Dynamic Model for Human Epidemics

A discrete dynamic model for human epidemics was developed in present study. The model included major parameters as transmission strength and its decline parameters, mean incubation period, hospitalization time, non-hospitalization daily mortality, non-hospitalization daily recovery rate, and hospitalization proportion, etc. Sensitivity analysis of the model indicated the total cumulative cases significantly increased with initial transmission strength, hospitalization time. The total cumulative cases significantly decreased with transmission strengths decline and hospitalization proportion, and linearly decreased with non-hospitalization daily mortality and non-hospitalization daily recovery rate. In a certain range, the total cumulative cases significantly increased with mean incubation period. Sensitivity analysis demonstrated that dynamic change of transmission strength is one of the most important and controllable factors. In addition, reducing the delay for hospitalization is much effective in weakening disease epidemic. Non-hospitalization recovery rate is of importance for enhancing immunity to recover from the disease.

microbiology↗
bioRxiv2020

Transient "rest" induces functional reinvigoration and epigenetic remodeling in exhausted CAR-T cells

T cell exhaustion limits immune responses against cancer and is a major cause of resistance to CAR-T cell therapeutics. Using a model wherein tonic CAR signaling induces hallmark features of exhaustion, we employed a drug-regulatable CAR to test the impact of transient cessation of receptor signaling (i.e. "rest") on the development and maintenance of exhaustion. Induction of rest in exhausting or already-exhausted CAR-T cells resulted in acquisition of a memory-like phenotype, improved anti-tumor functionality, and wholescale transcriptional and epigenetic reprogramming. Similar results were achieved with the Src kinase inhibitor dasatinib, which reversibly suppresses CAR signaling. The degree of functional reinvigoration was proportional to the duration of rest and was associated with expression of transcription factors TCF1 and LEF1. This work demonstrates that transient cessation of CAR-T cell signaling can enhance anti-tumor potency by preventing or reversing exhaustion and challenges the paradigm that exhaustion is an epigenetically fixed state.

immunology↗
bioRxiv2019

SIRT2-knockdown Rescues GARS-induced Charcot-Marie-Tooth Neuropathy

Charcot-Marie-Tooth disease is the most common inherited peripheral neuropathy. Dominant mutations in glycyl-tRNA synthetase (GARS) gene cause peripheral nerve degeneration and lead to CMT disease type 2D. Mutations in GARS (GARSCMT2D) show partial loss-of-function features, suggesting that tRNA-charging deficits play a role in disease pathogenesis, but the underlying mechanisms are not fully understood. In this study we report that wild-type GARS tightly binds the NAD+-dependent deacetylase SIRT2 and inhibits its deacetylation activity, resulting in the hyperacetylated -tubulin, the major substrate of SIRT2. Previous studies showed that acetylation of -tubulin protects microtubules from mechanical breakage and keep axonal transportation. However, CMT2D mutations in GARS can not inhibit SIRT2 deacetylation, which leads to decrease acetylated -tubulin and severe axonal transport deficits. Genetic reduction of SIRT2 in the Drosophila model rescues the GARS-induced axonal CMT neuropathy and extends the life span. Our findings demonstrate the pathogenic role of SIRT2-dependent -tubulin deacetylation in mutant GARS-induced neuropathies and provide new perspectives for targeting SIRT2 as a potential therapy against hereditary axonopathies.

cell biology↗
bioRxiv2019

Graph Convolutional Networks for Epigenetic State Prediction Using Both Sequence and 3D Genome Data

Predictive models of DNA epigenetic state such as transcription factor binding are essential for understanding regulatory processes and developing gene therapies. It is known that the 3D genome, or spatial structure of DNA, is highly influential in the epigenetic state. Deep neural networks have achieved state of the art performance on epigenetic state prediction by using short windows of DNA sequences independently. These methods, however, ignore the long-range dependencies when predicting the epigenetic states because modeling the 3D genome is challenging. In this work, we introduce ChromeGCN, a graph convolutional network for epigenetic state prediction by fusing both local sequence and long-range 3D genome information. By incorporating the 3D genome, we relax the i.i.d. assumption of local windows for a better representation of DNA. ChromeGCN explicitly incorporates known long-range interactions into the modeling, allowing us to identify and interpret those important long-range dependencies in influencing epigenetic states. We show experimentally that by fusing sequential and 3D genome data using ChromeGCN, we get a significant improvement over the state-of-the-art deep learning methods as indicated by three metrics. Importantly, we show that ChromeGCN is particularly useful for identifying epigenetic effects in those DNA windows that have a high degree of interactions with other DNA windows.

bioinformatics↗