For example, while we confirmed that the performance of DeepMod was comparable to that of other tools for the bacteria genome, DeepMod performed poorly at the human whole-genome scale, which was not previously reported.
Also, we showed that the performance of METEORE across the human genome was worse than that of Nanopolish, Megalodon, DeepSignal, and Guppy at the genome-wide scale across four datasets, which is distinctive from the evaluation by Yuen et al.
This is likely due to their small training dataset— CpG sites from E. Second, to compliment current publicly available benchmark datasets generated from human cell lines, we generated 1 two new nanopore sequencing datasets: one is from a primary leukemia specimen, and one is a human leukemia cancer cell line, and 2 WGBS and oxBS-seq datasets.
The new datasets enable the evaluation of these tools in a primary human specimen, not only in human cell lines, and thus will provide guidance on the application of nanopore sequencing in clinical research. In total, we used four human datasets with different coverages, which are large benchmark datasets than the prior benchmark studies.
Third, we evaluated the prediction robustness not only at a per-site level, but also at a per-read level, and considered more diverse genomic contexts, e. Fourth, we demonstrated that the 5hmC levels contribute to the discrepancy between BS-seq and nanopore sequencing. Fifth, we also compared the number of CpGs predicted by each tool and the computational resources consumed by each tool.
For example, the raw fast5 data from a single nanopore sequencing library, e. Thus, the consumption of computational resources is essential for guiding the design of data analyses on HPC and cloud computing platforms for large-scale human nanopore sequencing data.
Oxford Nanopore long-read sequencing technology poses both opportunities and challenges for accurate methylation prediction and long-range epigenetic phasing. The past few years have witnessed rapid development of both the sequencing technology and analytical tools. For DNA methylation analysis, many algorithms are emerging for nanopore sequencing data, and we comprehensively surveyed all current publicly available computational tools.
Based on our systematic comparison, we summarized the performances of seven tools across all major evaluation criteria Fig. We derived five key observations. First, the choice of methylation-calling tool critically affects the level of the F1 score, accuracy, and the AUC score at different genomic regions. DeepMod exhibited comparable performance for E. Second, detection of 5mCs at regions with discordant DNA methylation patterns, intergenic regions, low CG density regions, and repetitive regions i.
Therefore, penalized models, i. Third, Guppy and Nanopolish had the lowest memory usage, while Guppy, Nanopolish, and Megalodon are faster than all other tools. Fourth, we confirmed that the discrepancy in 5mC levels between the BS-seq and nanopore sequencing data results in part from the 5hmC modifications. Unlike nanopore sequencing, BS-seq cannot distinguish between 5mCs and 5hmCs, as bisulfite treatment does not convert either modification.
Thus, methylation calling using nanopore sequencing will benefit from future endeavors to increase the accuracy of challenging regions, the predicted CpG coverage, and high efficiency [ 91 ] in using computing resources.
Summary of per-read and per-site performances across all major evaluation criteria. Summary of A per-read performance F1 score , B per-site performance Pearson correlation coefficient , and resource usage. Details in evaluation criteria and cutoff values for performance categories are available in the Methods and Additional file 2 : Table S Therefore, we believe that our benchmarking of methylation-calling tools will guide researchers in making well-considered and effective choices when designing an analytic plan for epigenomic profiling using ONT sequencing, including Cas9-targeted nanopore sequencing data analysis.
For users with limited computational resources, we recommend Guppy and Nanopolish for methylation analysis. Guppy requires minimum CPU hours and peak memory as one of the top four performers because the base-modification prediction is part of its basecalling. Nanopolish is the best option considering per-read and per-site performance criteria, as well as low CPU hours and peak memory usage after basecalling. For users with the access to HPC resources or a larger budget for cloud computing resources, Megalodon is the best option, considering its performance in the more challenging areas including repetitive regions and discordant non-singletons, also as it predicts more CpG sites compared to Nanopolish and Guppy.
Robust prediction of DNA methylation at different genomic contexts will help improve our understanding of epigenetic mechanisms in gene regulation underlying many biological processes, including mammalian normal development, aging, and complex disease development.
In the study, we used four independent human datasets: two normal B-lymphocyte cell lines NA [ 56 ], NA [ 57 ] , one primary acute promyelocytic leukemia clinical specimen APL , and one cancer cell line K The Core maintains a tissue bank of cells from patients with hematologic malignancies. The patient sample was collected at the time of clinical presentation and prior to therapy. The sample was collected as leukapheresis and viably frozen using standard techniques.
We downloaded the v26 comprehensive genes annotation file gencode. All analyses were restricted to chromosomes , X, and Y.
Introns were generated by taking the difference between the genes file and the exons file. Intergenic regions were generated by taking the difference between the reference genome and all other gene feature types gene, CDS, promoter, intron from the gene annotation file using bedtools subtract. The sample destined for oxBS-seq was first subjected to oxidation whereas the sample destined for the WGBS library were mock-treated, and then followed by a bisulfite conversion.
The final library was quantified by real-time qPCR for an accurate concentration since proper quantitation is needed for loading the library for next-generation sequencing. For each dataset, we took the intersections of fully methylated or unmethylated CpG sites from BS-seq and those from all tools for per-read performance evaluation. Specially, if the estimated methylation level falls out of the confidence interval of binomial test calculated from input coverage and methylation level, then such event is counted as one conflict; the site is not reliable if more conflicts happen on one site [ 93 ].
The libraries were sequenced on Flow Cell R9. We obtained nanopore raw data and their library preparation details from the authors of [ 56 ]. The library was sequenced on Flow Cell R9. We downloaded the E. Basecalling, the process of translating raw electrical signal of nanopore sequencing into nucleotide sequence, is the initial step of nanopore data analysis. Both ONT and independent researchers are actively developing different tools for the basecalling step. Specifically, ONT provides basecalling programs including official ONT community-only software Albacore and Guppy and open-source software Flappie, Scrappie, Taiyaki, Runnie, and Bonito [ 95 ] , the latter of which are under development with new algorithms for basecalling.
Albacore [ 96 ] is a general-purpose base caller that runs on CPUs. Guppy [ 51 ] is a neural network based basecaller with several bioinformatic post-processing features. ONT discontinued to develop Albacore due to the better performance of Guppy [ 54 ]. Because the state-of-art basecaller Guppy using the default high-accuracy HAC model showed excellent performance among ONT basecalling tools [ 54 ], we used Guppy v4.
Specially, R9. We evaluated the performance of Nanopolish v0. These seven tools differ in the underlying algorithms and the modifications they are trained to detect DNA methylation.
Nanopolish groups nearby CpG sites together and calls the cluster jointly to assign the same methylation status to each site in the group. We used 2. Then we calculated per-site methylation frequency by the fraction of reads classified as methylated. Megalodon predict 5mC at either the per-read or per-site level by aggregating per-read results based on the log probability that the base is modified or unmodified. DeepSignal required an extra re-squiggle module of Tombo before methylation calling.
Guppy is a ONT-developed basecaller [ 51 ] and is able to identify certain types of modified basecalling i. Then, we used ONT-developed fast5mod v1. ONT-developed Tombo [ 20 ] performed a statistical test to identify modified nucleotides without the need for the training data.
Tombo computed per-read, per-site test statistics by comparing the signal intensity difference between modified bases and unmodified bases. The input is a reference genome and fast5 files with raw signals basecalled by Guppy v4.
The output is a BED file with coverage, number of methylated reads, and methylation percentage information for genomic positions of interest. Since a 5mCs in a CpG motif has a cluster effect in the human genome [ 34 ], DeepMod provides a cluster model to generate a final output for site-level-predicted methylation probability in the human genome.
Also, since DeepMod aggregated methylation calling results into a per-site output BED file, we counted the number of methylated callings and unmethylated callings from BED outputs to evaluate its read-level performance.
The performances of these methods that use prior knowledge about the expected deviations in signal are highly dependent on the training data, which is typically composed of a fully unmodified and a fully modified sample.
Motifs that are not represented in the training set or that contain mixtures of modified and unmodified bases may lead to suboptimal performance. We designed the performance-evaluation process for 5mC predictions among seven tools as follows. Second, we calculated the F1 score, accuracy, precision, and recall and assessed the tradeoff between true-positive and false-positive rates of 5mC predictions by calculating the ROC curve by varying the threshold for methylation calling and reported the AUC values as follows:.
We calculated F1 score for both 5mCs and 5Cs and used macro F1 score, i. AUC is a performance metric used to evaluate how well a classifier performs on both methylated and unmethylated class predictions.
We compared the memory usage and running time of the seven tools on the single-read fast5 files of each dataset. All tools have support for multi-processors, and we compare the scalability of these tools on the same system configurations. We split the ONT datasets for parallelization.
The HPC platform software and hardware specifications are as follows: slurm manager version: These results were used as the measurement of running time and memory usage for hardware performance comparison and evaluation. To facilitate the dissemination of DNA methylation calling results using nanopore sequencing from the current benchmark study, we present a web application named nanome.
Nanome is a user-friendly interactive nanopore sequencing methylation database and is implemented with Shiny package from R programming language. The database allows the users to select their features of interest, including chromosomes, strands, datasets, singletons and non-singletons, genomic contexts, regions of various CG density, and repeat regions. Nanome also provides methylation percentage and read coverage at each genome site across different methylation calling tools and bisulfite sequencing.
Figure 8 summarized the performance of each methylation-calling tool across the range of evaluation metrics. We calculated the F1 scores, Pearson correlation coefficients, CPU utilized time, and peak memory by the median value of each tool achieved across the four datasets.
ONT sequencing data for E. Nucleic acid modifications in regulation of gene expression. Cell Chemical Biology. Smith ZD, Meissner A. DNA methylation: roles in mammalian development. Nature Reviews Genetics. Profiling DNA methylation based on next-generation sequencing approaches: new insights and clinical applications. DNA N6-methyladenine: a new epigenetic mark in eukaryotes? Nat Rev Mol Cell Biol. Somatic mutations drive specific, but reversible, epigenetic heterogeneity states in AML.
Cancer Discovery. Coming of age: ten years of next-generation sequencing technologies. Nat Rev Genet. Advancements in next-generation sequencing. Ann Rev Genom Human Genet. Biosciences P. Detecting DNA base modifications using single molecule, real-time sequencing. Accessed 19 Sept Detecting DNA cytosine methylation using nanopore sequencing. Nat Methods. Yang Y, Scott SA. Methods Mol Biol. From squiggle to basepair: computational approaches for improving nanopore sequencing read accuracy.
Genome Biology. Improved data analysis for the MinION nanopore sequencer. Oxford Nanopore Technologies. Product comparison. Continuous development and improvement. De novo identification of DNA modifications enabled by genome-guided nanopore signal processing. Company history. Nanopore sequencing technology and tools for genome assembly: computational analysis of the current state, bottlenecks and future directions.
Brief Bioinform. On the application of BERT models for nanopore methylation detection. Structural and mechanistic insights into the bacterial amyloid secretion channel CsgG. Detection and mapping of 5-methylcytosine and 5-hydroxymethylcytosine with nanopore MspA.
Proc Natl Acad Sci. Carter J-M, Hussain S. Wellcome Open Res. R10 evaluation by GrandOmics the road to high accuracy of single nucleotide.
Flow Cell R Error rates for nanopore discrimination among cytosine, methylcytosine, and hydroxymethylcytosine along individual DNA strands. Proceedings of the National Academy of Sciences. BMC Genomics. Guppy protocol: modified base calling. Single-molecule sequencing detection of N6-methyladenine in microbial reference materials.
Nat Commun. Detection of DNA base modifications by deep recurrent neural network on Oxford Nanopore sequencing data. Nature Communications. Ni P, Huang N, others.
Systematic benchmarking of tools for CpG methylation detection from Nanopore sequencing. Mapping DNA methylation with high-throughput nanopore sequencing. Nature Methods. Li E, Zhang Y. DNA methylation in mammals. Cold Spring Harbor Perspectives in Biology. Almouzni G, Cedar H. Maintenance of epigenetic information.
The diverse roles of DNA methylation in mammalian development and disease. Megabase-scale methylation phasing using nanopore long reads and NanoMethPhase. Genome Biol. Opportunities and challenges in long-read sequencing data analysis.
Long-read whole-genome methylation patterning using enzymatic base conversion and nanopore sequencing. Nucleic Acids Res. Nondestructive enzymatic deamination enables single-molecule long-read amplicon sequencing for the determination of 5-methylcytosine and 5-hydroxymethylcytosine at single-base resolution.
Genome Res. Specifically, genetic variants in the spleen-specific promoters were enriched during Asian pig domestication, whereas variants within cortex-specific promoters were enriched during European pig domestication. This insight may reflect the observed distinct phenotypic difference between Asian more resistant to malaria 42 , 43 and European domesticated pigs more active and aggressive 44 , Further investigation is warranted to deepen our understanding of genetic selection and domestication in the pig.
This regulatory element atlas will serve as a valuable source for the livestock community to inform GWAS and eQTL findings, genomic selection programs, and genome editing strategies, as well as to enhance our understanding of genome evolution and adaptation.
With continued efforts by the FAANG Consortium 53 , more epigenomic data will be available from diverse samples, such as reproductive tissues, additional developmental stages, and different physiological states.
Finally, this atlas of functional elements provided a unique opportunity for comparative epigenomic analysis between human, mouse and pig, the results of which can inform which species constitute the most appropriate biomedical model s for specific human diseases.
We observed that regions under positive or negative selective pressure demonstrated higher conservation of epigenetic signatures such as TssA, TssBiv and TxFlnk than those which are not subject to selective pressure i.
Recently evolved liver enhancers i. Such enhancers have been demonstrated to actively affect gene expression, although they have a smaller effect than enhancers shared across species when the comparison is controlled for number of enhancer elements acting on a given gene However, human-specific promoters in brain tissues were enriched in intelligence-related genes, which suggests a critical role for epigenomic regulation of novel biological function in humans in the most evolutionarily conserved regions.
It is widely accepted that neither mouse nor pig is universally appropriate to serve as an animal model for every human disease 18 , Gene regulatory networks play significant roles in controlling phenotypic variance of complex traits, including most human diseases. In examining heritability enrichment of 47 complex traits in humans, our epi-conservation analysis among three species comparing pig-human vs. This line of evidence is consistent with many studies of human diseases using either mouse or pig as an animal model Our study provides a basis for understanding genetic regulation of complex traits, such as human diseases, by focusing on regulatory network conservation across different mammalian species.
Although the findings from our study are intriguing, experimental studies and more epigenomic data from additional tissues, cell types, and species — such as non-human primates — will be needed to extend and functionally validate the biological mechanisms that underpin complex traits and diseases 9 , Five gut-associated tissues stomach, jejunum, duodenum, ileum, and colon were collected from two Yorkshire littermate male pigs at six months of age from Michigan State University Cecum from two female hybrid pigs Yorkshire-Hampshire cross, five months of age were obtained at University of California, Davis meat laboratory.
C, Denville, NJ , as previously described An input no antibody was performed for each sample. Briefly, the susScr11 genome assembly and Ensembl genome annotation v were used as references for pig.
Sequencing reads were trimmed with Trim Galore! For RNA-seq, gene counts were determined using htseq-count 63 v.
For ChIP-seq, after the filtering, duplicates were marked and removed using Picard v. Various quality metrics e. RRBS data were processed using Bismark 66 v. Hi-C contacts were called using the Juicer pipeline 67 with default parameters. The global correlations among assays, tissues, and biological replicates were performed by deepTools 68 v. The signal of marks along with protein-coding genes were generated by deepTools 68 computeMatrix scale-regions function with parameters -a -b The Z -score was used to normalize bigWig of five marks given input files.
ChromHMM 69 v. The same tissue of two biological replicates were collectively considered as one tissue epigenome. The state model was chosen, as it presented maximum number of states with distinct epigenetic mark combinations. We labeled these 15 chromatin states based on their combinations of histone modifications and enrichment around TSS 8 , Then the fold enrichment of each chromatin state for each external gene element e.
M3 was used as a reference, since it was closest to median expression. States whose cumulative coverage changed faster than others were considered to be less constitutive more variable states.
Chromatin state switching between tissues was calculated by pairing two tissues. By averaging these calculations for a pair of tissues, we obtained the pair switching probabilities. We calculated the state switching probabilities in between intestinal tissues, between brain tissues Supplementary Fig. For example, we extracted H3K4me1 signal confidence scores for EnhA. We first grouped some tissues into different sub-groups, such as small intestine jejunum, ileum, and duodenum , large intestine cecum, colon , and brain cortex, cerebellum, and hypothalamus.
Then we scaled the log2-transformed expression i. Further we computed a t-statistic to identify tissue-specific expressed genes by excluding the tissues in the same sub-group. Several other methods could be also used to detected tissue-specific genes To evaluate how enhancers of TSE genes switch among tissues, we first identified the target enhancers of TSE genes following the method described in our recent study Finally, we computed enhancer state switching probabilities of TSE genes among tissues using the method described above.
For strong enhancers EnhA identified in each tissue, we counted the bins of overlapping RRATs by comparing to other tissues. We generated a total of 17 modules of tissue-specific regulatory element TSR enhancers.
These 17 modules included all-common presented in all tissues , gut-common presented in all 5 intestinal tissues , brain-common presented in all 3 brain tissues and 14 tissue-specific modules.
We selected the top three enriched or tissue function-relevant motifs for each tissue as the candidate tissue-specific EnhAs motifs and identified a total of 51 motifs enriched in tissue-specific EnhAs.
The mRNA expression of corresponding TFs in pigs were used to calculate the correlation with motif enrichment. A total of whole genome sequence WGS datasets Supplementary Data 9 in pigs Asian wild 58 and domestic pigs , European wild 35 and domestic pigs were trimmed by Trimmomatic 79 v. All genome variant calls were then combined and the variants for each sample were called by GenotypeGVCFs.
The pig GWAS data of 44 traits was described previously , Briefly, more than , pigs Supplementary Data 12 were genotyped by a variety of Porcine chip arrays 8. Then these genotyped animals were imputed to genome-wide level using an intermediated reference panel of animals genotyped by a K SNP array, then a reference panel of WGS datasets.
Furthermore, we filtered out all SNPs with either 1 a minor allele frequency below 0. Last, we performed GWAS signal enrichment of 44 pig complex traits 3 ADG-related, 20 lipid-related, and 21 feed efficiency-related for each chromatin state across 14 tissues by applying a genotype cyclical permutation test, repeated 10, times In total, we obtained six matched tissues small intestine, liver, spleen, lung, adipose, brain cortex among pig, human, and mouse.
All data were processed following the same pipeline used in pig. Chromatin states of human and mouse were also trained by ChromHMM and 15 chromatin states were identified. To explore the relationship between sequence conservation and epi-conservation among the three mammals, we first divided the genome into 50 equally sized sets 0thth with increasing average PhyloP scores using the method detailed by Xiao et al. These genomic segments were divided into 50 equally sized sets from the fastest changing sequence smallest PhyloP scores to the most conserved greatest PhyloP scores.
Supplementary Fig. All chromatin states in pig and mouse were lifted over to human. The conservation rate 0—1 of each region of each state from pig to human was calculated based on state region coverage of pig over human. If there was no overlap it was assigned 0, if completely occupied it was assigned 1. The same analysis was conducted for pig to mouse and mouse to human.
Furthermore, we performed genomic and epigenomic conservations for every pair of mammalian species in each tissue. To examine the biological relevance of regions with extremely variable sequence 0—2th sets or highly conserved sequence 47—49th sets , we extracted the human-pig shared and human-specific chromatin state TssA from these sets.
We sorted the genes by p-value within each species and divided them into 50 equally sized sets. We then performed heritability enrichment analysis by applying stratified linkage disequilibrium score regression LDSC to partition heritability of 47 human complex traits into distinct functional categories Stratified linkage disequilibrium score regression LDSC is a commonly used approach to partition the heritability of functional annotations and to estimate the enrichment degree i.
In this study, LDSC was used to determine the SNP-based heritability estimates, and then partition the heritability into separate functional categories to demonstrate the disproportionate contribution of different functional categories to the heritability of human complex traits and diseases.
These functional categories included six types of species-shared and species-specific regulatory elements, chromatin states of each tissue, and TSR of EnhA and TssA.
The GWAS summary statistics for 47 human complex traits were obtained from public databases Supplementary Data 16 , with an average sample size of , all European ancestry and a high-quality overlap with HapMap3 panel. The results of LDSC regression for the base model, which has not been partitioned for heritability, are available in Supplementary Data Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Consortium, E. Expanded encyclopaedias of DNA elements in the human and mouse genomes. Nature , — Science , — Nature , Dunham, I. An integrated encyclopedia of DNA elements in the human genome. Nature , 57 Filion, G. Systematic protein location mapping reveals five principal chromatin types in Drosophila cells. Cell , — Ernst, J.
Mapping and analysis of chromatin state dynamics in nine human cell types. Nature , 43—49 Pang, B. Systematic identification of silencers in human cells.
Roadmap Epigenomics, C. Integrative analysis of reference human epigenomes. Gorkin, D. An atlas of dynamic chromatin landscapes in mouse fetal development. Zabidi, M. Enhancer—core-promoter specificity separates developmental and housekeeping gene regulation. He, Y. Spatiotemporal DNA methylome dynamics of the developing mouse fetus. Roy, S. Gerstein, M. Maurano, M. Systematic localization of common disease-associated variation in regulatory DNA.
Boix, C. Regulatory genomic circuitry of human disease loci by integrative epigenomics. Zhang, Q. A pig model of the human gastrointestinal tract.
Gut Microbes 4 , — Bassols, A. The pig as an animal model for human pathologies: a proteomics perspective. Proteomics Clin. Meurens, F. The pig: a model for human infectious diseases. Trends Microbiol. Sullivan, T. The pig as a model for human wound healing. Wound Repair Regen. Gieling, E. Molecular and Functional Models in Neuropsychiatry. Kragh, P. Transgenic Res. Pig proteomics: a review of a species in the crossroad between biomedical and food sciences. Xiang, R. Quantifying the contribution of sequence variants with regulatory and evolutionary significance to 34 bovine complex traits.
Natl Acad. USA , — Andersson, L. Genome Biol. Burns, E. Generation of an equine biobank to be used for Functional Annotation of Animal Genomes project. Kingsley, N. Functionally annotating regulatory elements in the equine genome using histone mark chip-seq. Genes 11 , 3 Fang, L. Functional annotation of the cattle genome through systematic discovery and characterization of chromatin states and butyrate-induced variations.
BMC Biol. Halstead, M. A comparative analysis of chromatin accessibility in cattle, pig, and mouse tissues. BMC Genomics 21 , 1—16 Colin Kern, Y. Functional genome annotations of three domestic animal species provide a vital resource for comparative and agricultural research.
Foissac, S. Multi-species annotation of transcriptome and chromatin structure in domesticated animals. Tyska, M. Myosin-1a is critical for normal brush border structure and composition. Cell 16 , — Shifrin, D. Jr et al. Enterocyte microvillus-derived vesicles detoxify bacterial products and regulate epithelial-microbial interactions. Wagner, J. Our results indicate that, under chronic hypoxia, lower levels of Tmod3 play an important role in the maintenance or neo-vascularization of pulmonary arteries.
Most users should sign in with their email address. If you originally registered with a username please use that to sign in. To purchase short term access, please sign in to your Oxford Academic account above. Don't already have an Oxford Academic account? Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide. Sign In or Create an Account. Sign In. Advanced Search.
Search Menu. Article Navigation. Close mobile search navigation Article Navigation. Article Contents Abstract. Materials and Methods.
Corrected proof. Heterozygous Tropomodulin 3 mice have improved lung vascularization after chronic hypoxia. Tsering Stobdan , Tsering Stobdan. Division of Respiratory Medicine. Oxford Academic. Google Scholar.
0コメント