Predicting the past and future evolutionary space of SARS-CoV-2

doi:10.21203/rs.3.rs-3195284/v1

Download PDF

Biological Sciences - Article

Predicting the past and future evolutionary space of SARS-CoV-2

https://doi.org/10.21203/rs.3.rs-3195284/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

SARS-CoV-2 emerged in the human population in late 2019^1,2, went through population bottlenecks³ and has undergone adaptation and evolution by single nucleotide polymorphisms and recombination. In individual humans, viral populations consist of dominant (the most common) and minor variant genomes⁴. Subsequent spread has driven the selection of variants that are more transmissible and capable of partial immune escape^5-7. However, whether novel variants can be predicted is unknown. An analysis of 96,559 SARS-CoV-2 genomes and their associated variation in samples sequenced over a three-year period paint a detailed picture of viral population genetics. Variants contained non-synonymous amino acid changes, but there was little synonymous variability. Non-synonymous changes have increased in frequency as the pandemic progressed, with spike having the most changes. Amino acid variation associated with the Alpha and Delta Variants of Concern (VoCs) took three months to replace the previous lineages, however this occurred more quickly with Omicron. Mutational signals associated with VoCs and the current Omicron lineage were observed near the start of the pandemic in the UK, suggesting it was only a matter of time before such VoCs emerged. Machine leaning was used to develop a model that predicted the variable amino acid sites in spike that may be subject to future mutation and thus can inform design of medical countermeasures and provide a risk-based evaluation of viral properties.

Biological sciences/Microbiology/Virology/SARS-CoV-2

Biological sciences/Computational biology and bioinformatics

The mutational landscape of SARS-CoV-2 varies at the dominant viral genome sequence and minor variant genomes^4,8. After spill overs from the intermediate animal host in the Huanan Seafood Wholesale Market^1,2, subsequent human-to-human infection has resulted in SARS-CoV-2 genomic diversification from the original zoonotic spill overs. These mutations have arisen through genomic changes from virus and host derived single nucleotide polymorphisms (SNPs), homologous and heterologous recombination events, and insertions and deletions (indels). The first major changes in the SARS-CoV-2 genome in the human population were the D614G substitution in the spike glycoprotein (spike_D614G) accompanied by the P323L substitution in the viral RNA dependent RNA polymerase (NSP12) (NSP12_P323L). These changes were associated with increased transmissibility, replication and fitness^5,6,9.

Diversity in dominant and through minor variant genomes exists in humans/animals infected with SARS-CoV-2¹⁰ (Extended Data Fig. 1A). These genomes will have both synonymous (non-coding) and non-synonymous (coding) differences. Minor variant genomes may become dominant when they confer a selective advantage either through the specific repertoire of SNPs and/or recombination. Since the start of the COVID-19 pandemic, different dominant viral genome sequences and non-synonymous changes appear to rise and fall in the SARS-CoV-2 global sequences^11,12. Genetic change has resulted in the sequential dominance of several Variants of Concern (VoCs) associated with waves of high case numbers throughout the pandemic. With time, VoCs have emerged with examples representing large genomic jumps from circulating strains, with many novel mutations having the appearance of occurring at once, such as with the emergence of Alpha¹³ and Omicron¹⁴. One hypothesis is that these divergent variants may stem from persistent infection in immune-deficient hosts^13,15. In this study, artificial intelligence/machine learning (AI/ML) was used to investigate the evolutionary patterns of amino acids changes at a population level and used to predict the conserved and future mutable amino acid sites of the spike protein, informing vaccine and antibody therapies for SARS-CoV-2.

Genetic variation in SARS-CoV-2 increased with time.

To understand the emergence of minor variant genomes, and derive amino acid variation frequencies, sequence data gathered by COG-UK (COVID-19 Genomics UK consortium) was analysed. Sequences produced by the ARTIC amplification method and sequenced by Illumina NovaSeq 6000 were selected. In total, 96,559 SARS-CoV-2 consensus genomes were derived, together with the associated minor variant genomic information (representing approximately 100 genomes per day for nearly three years, starting in March 2020 up to 31^st December 2022). The nucleotide sequences were in silico translated into amino acid sequences. The average variation frequencies of each amino acid were calculated to investigate the pattern of mutational emergence in SARS-CoV-2 proteins over the three-year period. We found the presence of artificial low-frequency genetic variants due to errors acquired during amplicon generation and sequencing¹⁶ (Extended Data Fig. 1B and C; see more details in Supplementary File 1). To mitigate this effect, the average variation frequencies of amino acids at the same site in different months was compared with each other to reveal differences in frequencies of amino acid changes, instead of comparing different sites. The average variation frequencies of amino acid sites in each protein were compared across time by month (Fig. 1a). All proteins (except ORF7a, ORF7b, ORF8 and ORF10) had a similar pattern of increasing diversity from the start of the pandemic until a slight decrease associated with the bottleneck and spread of the emergence of the Alpha variant in October 2020 and then subsequent increasing diversity at the minor genomic variant level between December 2020 and January 2021.

Emergence of amino acid substitutions associated with VoCs in the UK.

The monthly average variation frequency of each amino acid site in SARS-CoV-2 from April 2020 to October (1st to 15th) 2021 was compared to that of March 2020 (Extended Data Fig. 2 – animation of variation for each protein). Because the ARTIC primers were updated from version 3 to version 4 from 15th October 2021 (to mitigate amplicon drop-outs caused by the emergence of the Delta variant), the monthly average variation frequency from November 2021 to December 2022 was compared to that of October (16th to 31st) 2021. As a proportion of the viral genome/proteome, the data indicated that spike had the greatest variation, reflecting the fact that spike is under strong selection pressure from host immunity (Extended Data Fig. 2). As an example, the average variation frequencies of amino acid sites in April 2020 all data points revolved around the 1:1 line, indicating that the variation frequencies of amino acid sites in April were like those in March for spike (Extended Data Fig. 3A). The yellow-colored dot in the figure that separated from the major cluster is the D614G substitution with average variation frequency ~1.0 in either March or April 2020⁵. The D614G substitution was then maintained in all subsequent variants in this study without a decline in the variation frequency over a three-year period (Extended Data Fig. 3).

Two variants (Alpha and Delta) were prevalent in the UK before the introduction and evolution of the Omicron variants (Extended Data Fig. 3). Mutations of Alpha were emerging from October 2020 and were fixed in January 2021 (Fig. 1B and Extended Data Fig. 3). Afterwards, mutations of Delta emerged from April 2021 (Fig. 1C and Extended Data Fig. 3) and were fixed by June 2021, whilst Alpha and Delta were coexisting in May 2021. Our previous findings showed that it took three months for the D614G mutation (associated with the Lineage B.1 variant) to increase in frequency from the minor variant level to the dominant sequence during the containment phase in the UK with the two variants coexisting in the same sample (and therefore person) at a high frequency in the second month⁵. Most of the mutations from the Alpha and Delta variants followed the same pattern of transition in the analyzed data, except P681H from the Alpha variant (Fig. 2A and Extended Data Fig. 3), and R158G, A222V and P681R from the Delta variant (Fig. 2B and Extended Data Fig. 3). P681H (Alpha) and R158G (Delta) became dominant sequence at a slower rate than other amino acid substitutions associated with these VoCs. Meanwhile, P681R (Delta) was maintained at a constant coexisting frequency (note that for Delta there were several co-circulating sub-lineages¹⁷). Substitutions associated with the first Omicron variant emerged from the minor variant genomes to the dominant sequence from December 2021 and became fixed by February 2022 (Extended Data Figs. 2 and 3). Afterwards, new mutations were continually emerging and withdrawing, associated with new Omicron sub-variants (Extended Data Figs. 2 and 3).

Evolutionary patterns and mutational signals from the start of the pandemic

To investigate whether we could predict or rank potential amino acids in spike that would be subject to evolutionary pressure, the W value (referred to as the Shapiro-Wilk statistic W - a test of normality) was used to measure the evolutionary patterns of amino acids in these samples. If we assume that the virus was not subjected to any evolutionary forces, the observed variations at amino acid sites would be a result of amplification and sequencing errors and/or stochastic errors during viral replication, in which case the frequencies of variations at an amino acid site in monthly samples should exhibit symmetry around their mean and display an approximate normal distribution and have a high W value (Fig. 3A, blue curve). These amino acids would cluster to the top when the relationship between the W value and average variation frequency is visualized (amino acids contained within the blue circle in Fig. 3B). Viruses under evolutionary pressure would have increased frequencies of variation in some individuals (samples) at amino acid sites relevant to adapting to the pressure. This would increase the thickness of the right tail, generating a right-skewed distribution and lower W value (Fig. 3A, green curve). This would result in amino acids that were present in the top blue cluster shifting downwards - such amino acid sites would appear in the green circle of Fig. 3B. Stronger evolutionary pressure would increase the density of higher variation frequencies, generating a left-skewed distribution and a much lower W value (Fig. 3A, red curve). This would result in the amino acid site shifting to the far right of the plot, as highlighted in the red circle of Fig. 3B (further details can be found in Supplementary File 1).

The relationship between the W value and average variation frequency was visualised for each amino site in spike for each month. Two separated clusters for amino acid variation in spike were observed throughout all the months analyzed (Extended Data Fig. 5 and 6 and Fig. 3B). Some amino acids in spike kept high W values close to 1 through all three years of the pandemic and some shifted between high and low W value clusters (Extended Data Figs. 5 and 6). All amino acid substitutions that defined the VoCs travelled through the cluster associated with a low W value before they became fixed in the population (Extended Data Figs. 6). However, we found that amino acid sites associated with the VoCs presented in the low W value cluster or shifted between the two clusters (under selection pressures) from start of the pandemic (Extended Data Figs. 5 and 6). Especially, substitutions associated with Omicron could be observed in the low W value cluster in the first three months of UK data analyzed in 2020, although this variant emerged in late 2021 with a proposed origin in Southern Africa¹⁴ (Extended Data Figs. 5 and 6). These mutational signals suggest that such variants could have emerged anywhere and was perhaps only a matter of time and number of people infected (assuming the same pattern of recombination that led to VoCs).

Predicting past and future mutable sites in SARS-CoV-2 spike

The determination of a W value for each amino acid in spike in a large data set, given the background context of evolution in a population, may provide a predictor for whether that site would be subject to future selection. To predict which sites in spike may be subject to future evolutionary pressure we used an AI/ML technique that identified models (patterns, structures, or relationships) in a dataset without any prior knowledge. The K-Medoid Clustering (pam) algorithm was used to separate the conserved amino acid sites from the mutable amino acid sites in spike using the W values of each amino acid site as variables by month. With this algorithm, amino acid sites with similar W values were grouped into the same cluster based on the W values from March 2020 to May 2020 for three months (pam3), March 2020 to August 2020 for six months (pam6), March 2020 to August 2020 for nine months (pam9), March 2020 to February 2021 for 12 months (pam12) and March 2020 to December 2022 for 34 months (pam34) (Fig. 4A and B, Extended Data Fig. 7).

To evaluate the performance of the models to cluster amino acid with similar W values and to exam whether these were related to potential variability we examined substitutions in spike associated with WHO VoCs (https://covariants.org updated on 2023-04-26, Supplementary information 1), substitutions that are most frequent (tracked by https://www.bv-brc.org updated on 2023-01-12) in frequencies of 0.001 and 0.005, (Supplementary information 2) and substitutions that appeared in any Pango lineage (collected by https://cov-lineages.org extracted on 2023-06-10, Supplementary information 3). In general, the pam clusters performed better when the models were calculated using W values from more months (Fig. 4, Extended Data Fig. 7 and Table 1). Amino acid sites within the cluster exhibiting high W values are potentially conserved, while substitutions are more likely to occur on those amino acids within the cluster with low W values. In all the developed models, only a small number of tested mutation sites were found in cluster 1, which had the highest W values. Most of the tested mutation sites were found in the cluster with the lowest W values (Extended Data Table 1). Using the pam12 model, approximately 13% of the mutation sites of interest were found in cluster 1, while approximately 87% were found in the other clusters (Extended Data Table 1). In other words, the pam12 model correctly clustered around 87% of the mutation sites. The pam12 model successfully predicted 27 out of 39 (69.2%) Omicron-specific mutation sites that emerged since late 2021 (Supplementary Information 1). Similarly, the pam34 model correctly clustered 98% of the mutation sites of interest (Extended Data Table 1). Both the pam12 and pam34 models also demonstrated excellent performance in clustering the mutation sites that occurred most frequently (0.001 and 0.005) (Extended Data Table 2).

These models were also applied to cluster the 413 substitution sites in spike found across all Pango lineages (total 2600), including 306 mutation sites (in 2560 lineages) that were identified since March 2020 (Supplementary Information 3). Although most of these Pango lineages were not prevalent and the mutations occurred at these sites were rare, the pam12 model was still able to predict approximately 75% of the 306 mutation sites. This method not only clustered amino acid sites with similar patterns of W values but also served as a good predictor for identifying amino acid sites with a high likelihood of substitution. The conserved amino acid sites identified by the pam12 model included 92% (587) of the conserved amino acid sites identified by the pam34 model, indicating that W values derived using 12 months from start of the pandemic could be utilized to develop an effective predictive model for future evolutionary sites in spike.

The pam34 spike model was developed based on a large dataset (96,209 samples) collected over a three-year period since the beginning of the pandemic. The virus has been subjected to various evolutionary forces that are present in these individuals (e.g. transmission, prior infection with SARS-CoV-2 and thus immunity, vaccination). With the exception of a virological process such as recombination and treatment with antivirals, it is unlikely that many additional evolutionary forces will act on the virus at a population level. Furthermore, no substitutions caused by the vaccine have been reported¹⁸. Therefore, we predict that the pam34 model would be expected to remain applicable for a significant period in the future. The sites were modelled onto the spike protein using PyMOL (Fig. 4C and D) and, for example, identifies position V483 on the receptor binding domain that is associated with escape to monoclonal antibody therapy^19,20 and G486 that is associated with a loss in neutralisation²¹. This study identifies sites in the spike protein that are subject to change but also highlights those sites that remain conserved (Fig. 4C and D) and would logically inform part of a universal vaccine that would be protect against potential future variants.

Data collection

From COG-UK sequencing data 100 samples per day (where appropriate) were randomly selected with a fastq size larger than 100 Mb, spanning from March 2020 to December 2022. A total of 96,558 samples were used for this analysis (Supplementary Information 4). After removing the samples with a consensus genome covering less than 90% of the coding region, 96,209 samples were used for amino acid variation analysis (Supplementary Information 5). The fastq data were trimmed and cleaned by COG-UK. Therefore, we did not apply any further processing to these paired-end fastq files after retrieving them from the NCBI SRA database using "fastq-dump" in the SRA-Toolkit.

Amino acid variation analysis

The minor variations of amino acid in the genes of virus were called based on our previous methodologies^22,23. The paired-end fastq files obtained from NCBI SRA database were then mapped on a known SARS-CoV-2 genome (Wuhan-Hu-1, GenBank sequence accession: NC_045512.2) using Bowtie2 v2.3.5.1²⁴ by setting the options to parameters “--local -X 500 --no-mixed”, followed by SAM file to BAM file conversion, sorting, and removal of the reads with a mapping quality score below 11, not in pair, and not primary and supplementary alignment using SAMtools v1.9²⁵. Bamclipper (v 1.0.0)²⁶ was used to trim the ARTIC primer sequences on the mapped reads within the BAM file. The reads without ARTIC primer sequences were also excluded in the further analysis. The resultant Bam file was processed by Quasirecomb v1.2²⁷ to generate read coverage of each nucleotide site and a phred-weighted table of nucleotide frequencies. The phred-weighted table was parsed with a custom perl script to generate a dominant genome sequence as our previous description²⁸. The dominant genome sequence was then used as a template in the second round of mapping to generate a BAM file and consensus genome. Only samples with a consensus genome covering 90% of the coding region were further processed. This BAM file was then processed by the diversiutils script in DiversiTools (http://josephhughes.github.io/DiversiTools/) with the “-orfs” function to generate the coverage and amino acid variation at each amino acid position in the protein. Then, the number of amino acid changes caused by nucleotide deviations at each site, compared to the reference SARS-CoV-2 genome (NC_045512.2), was calculated. In order to distinguish low frequency mutations from Illumina sequence errors, the diversiutils script used the calling algorithms based on the Illumina quality scores to calculate a P-value for each mutation at each nucleotide site²⁹. The amino acid change was then filtered based on the P-value (< 0.05) to remove the low frequency mutations caused by Illumina sequence errors. This pipeline has been packaged into an efficient and user-friendly tool called AioMinor (https://github.com/Hiscox-lab/AioMinor) for processing large datasets in this project.

Shapiro-Wilk test and Skewness

The amino acid site variation frequency was calculated as the ratio between the number of amino acid changes and the coverage at each amino acid site. The average site amino acid variation frequency was calculated for each site per month. Only the amino acid sites with a coverage greater than 5 were included in the calculation. The normality of amino acid variation frequencies at each site in each month was tested using the Shapiro-Wilk test³⁰. The Shapiro-Wilk test was performed using the Perl module "Statistics::Normality", and the W value was reported as a measure of the normality of the test data³⁰. The asymmetry of the amino acid variation frequency distribution at each site in each month was measured by Skewness³¹, which was calculated by the Perl module "Math::Stat".

Unsupervised machine learning

Unsupervised machine learning was performed with R language v4.1.3³². The W values of each amino acid site in the selected months were combined into a data matrix. The PCA method was then applied to the data matrix to create new variables using the pca function in the PCAtools package (v2.6.0). The parameters "removeVar = 0.1, center = T, scale = T" in the pca function were used to standardize the data and remove 10% of variables based on low variance. Parallel analysis was carried out with the parallelPCA function from the PCAtools package to determine the number of principal components to retain. The retained principal components were used for clustering analysis using the pam algorithm from the cluster package (v2.1.4) with default settings. The optimal number of clusters in the pam algorithm was determined using the wss statistic from the fviz_nbclust function in the factoextra package (v1.0.7).

Data availability

The SARS-CoV-2 genomes sequencing data from COG-UK are currently freely available at the NCBI SRA database. Clusters in the models of pam3, pam6, pam9, pam12 and pam34 are available in Supplementary Information 6.

Code availability

The pipeline for deriving amino acid variation has been packaged into an efficient and user-friendly tool called AioMinor (https://github.com/Hiscox-lab/AioMinor) for processing the large datasets in this project.

Acknowledgements

This work was funded by U.S. Food and Drug Administration Medical Countermeasures Initiative contract (75F40120C00085). The article reflects the views of the authors and does not represent the views or policies of the FDA.  This work was supported by the MRC (MR/W005611/1) G2P-UK: A national virology consortium to address phenotypic consequences of SARS-CoV-2 genomic variation and the MRC funded G2P2-UK. JAH, MWC and IDB are supported by the NIHR Health Protection Research Unit (HPRU) in Emerging and Zoonotic Infections at University of Liverpool in partnership with the UK Health Security Agency (UK-HSA), in collaboration with Liverpool School of Tropical Medicine and the University of Oxford (award 200907). We would like to thank members of COG-UK for sequencing efforts during the COVID-19.

Contributions

Conceptualization: XD, TMac, TM, RO and JAH. Methodology: XD, GG, IDB, HG, AD and JAH. Validation: XD, AD, GG and DAM. Formal analysis: XD and JAH. Investigation: XD, GG and JAH. Resources: AD and JAH. Data curation: XD. Writing - Original Draft: XD and JAH. Writing - Review & Editing: all authors. Visualization: XD, IDB, HG and JAH. Supervision: AD, DB and JAH. Project administration: TMac, TM, RO and JAH. Funding acquisition: DAM, AD, MWC, DB and JAH.

Pekar, J. E. et al. The molecular epidemiology of multiple zoonotic origins of SARS-CoV-2. Science 377, 960-966, doi:10.1126/science.abp8337 (2022).
Worobey, M. et al. The Huanan Seafood Wholesale Market in Wuhan was the early epicenter of the COVID-19 pandemic. Science 377, 951-959, doi:10.1126/science.abp8715 (2022).
Lu, R. et al. Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding. Lancet 395, 565-574, doi:10.1016/S0140-6736(20)30251-8 (2020).
Moore, S. C. et al. Amplicon-Based Detection and Sequencing of SARS-CoV-2 in Nasopharyngeal Swabs from Patients With COVID-19 and Identification of Deletions in the Viral Genome That Encode Proteins Involved in Interferon Antagonism. Viruses 12, doi:10.3390/v12101164 (2020).
Goldswain, H. et al. The P323L substitution in the SARS-CoV-2 polymerase (NSP12) confers a selective advantage during infection. Genome Biol 24, 47, doi:10.1186/s13059-023-02881-5 (2023).
Plante, J. A. et al. Spike mutation D614G alters SARS-CoV-2 fitness. Nature 592, 116-121, doi:10.1038/s41586-020-2895-3 (2021).
Willett, B. J. et al. SARS-CoV-2 Omicron is an immune escape variant with an altered cell entry pathway. Nat Microbiol 7, 1161-1179, doi:10.1038/s41564-022-01143-7 (2022).
Prince, T. et al. Analysis of SARS-CoV-2 in Nasopharyngeal Samples from Patients with COVID-19 Illustrates Population Variation and Diverse Phenotypes, Placing the Growth Properties of Variants of Concern in Context with Other Lineages. mSphere 7, e0091321, doi:10.1128/msphere.00913-21 (2022).
Zhang, L. et al. SARS-CoV-2 spike-protein D614G mutation increases virion spike density and infectivity. Nat Commun 11, 6013, doi:10.1038/s41467-020-19808-4 (2020).
Lythgoe, K. A. et al. SARS-CoV-2 within-host diversity and transmission. Science 372, doi:10.1126/science.abg0821 (2021).
Worobey, M. et al. The emergence of SARS-CoV-2 in Europe and North America. Science 370, 564-570, doi:10.1126/science.abc8169 (2020).
Obermeyer, F. et al. Analysis of 6.4 million SARS-CoV-2 genomes identifies mutations associated with fitness. Science 376, 1327-1332, doi:10.1126/science.abm1208 (2022).
Hill, V. et al. The origins and molecular evolution of SARS-CoV-2 lineage B.1.1.7 in the UK. doi:10.1101/2022.03.08.481609 (2022).
Viana, R. et al. Rapid epidemic expansion of the SARS-CoV-2 Omicron variant in southern Africa. Nature 603, 679-686, doi:10.1038/s41586-022-04411-y (2022).
Kemp, S. A. et al. SARS-CoV-2 evolution during treatment of chronic infection. Nature 592, 277-282, doi:10.1038/s41586-021-03291-y (2021).
Guenay-Greunke, Y., Bohan, D. A., Traugott, M. & Wallinger, C. Handling of targeted amplicon sequencing data focusing on index hopping and demultiplexing using a nested metabarcoding approach in ecology. Scientific reports 11, 19510 (2021).
Vöhringer, H. S. et al. Genomic reconstruction of the SARS-CoV-2 epidemic in England. Nature 600, 506-511 (2021).
Gu, H. et al. Within-host genetic diversity of SARS-CoV-2 lineages in unvaccinated and vaccinated individuals. Nature Communications 14, 1793 (2023).
Laurini, E., Marson, D., Aulic, S., Fermeglia, A. & Pricl, S. Molecular rationale for SARS-CoV-2 spike circulating mutations able to escape bamlanivimab and etesevimab monoclonal antibodies. Scientific Reports 11, 20274 (2021).
Yi, C. et al. Key residues of the receptor binding motif in the spike protein of SARS-CoV-2 that interact with ACE2 and neutralizing antibodies. Cellular & molecular immunology 17, 621-630 (2020).
Greaney, A. J. et al. Comprehensive mapping of mutations in the SARS-CoV-2 receptor-binding domain that affect recognition by polyclonal human plasma antibodies. Cell host & microbe 29, 463-476. e466 (2021).
Dong, X. et al. Variation around the dominant viral genome sequence contributes to viral load and outcome in patients with Ebola virus disease. Genome biology 21, 1-20 (2020).
Goldswain, H. et al. The P323L substitution in the SARS-CoV-2 polymerase (NSP12) confers a selective advantage during infection. Genome biology 24, 47 (2023).
Kim, D., Langmead, B. & Salzberg, S. L. HISAT: a fast spliced aligner with low memory requirements. Nature methods 12, 357 (2015).
Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics 25, 2078-2079 (2009).
Au, C. H., Ho, D. N., Kwong, A., Chan, T. L. & Ma, E. S. BAMClipper: removing primers from alignments to minimize false-negative mutations in amplicon next-generation sequencing. Scientific reports 7, 1-7 (2017).
Töpfer, A. et al. Probabilistic inference of viral quasispecies subject to recombination. Journal of Computational Biology 20, 113-123 (2013).
Carroll, M. W. et al. Temporal and spatial analysis of the 2014–2015 Ebola virus outbreak in West Africa. Nature 524, 97 (2015).
Morelli, M. J. et al. Evolution of foot-and-mouth disease virus intra-sample sequence diversity during serial transmission in bovine hosts. Veterinary research 44, 12 (2013).
Royston, J. P. An extension of Shapiro and Wilk's W test for normality to large samples. Journal of the Royal Statistical Society: Series C (Applied Statistics) 31, 115-124 (1982).
Joanes, D. N. & Gill, C. A. Comparing measures of sample skewness and kurtosis. Journal of the Royal Statistical Society: Series D (The Statistician) 47, 183-189 (1998).
Team, R. C. R: A language and environment for statistical computing. (2013).

There is NO Competing Interest.

Supplementaryfile1.docx
Supplementary text
DongssonataofSARS2spike.unk
Audio file for Extended Data Fig. 6
EXTENDEDDATA.docx
Extended Data for figures
Supplementaryfigures.docx
Supplementary figures
Supplementarytables.xlsx
Supplementary tables

Download PDF

Version 1

posted

You are reading this latest preprint version

Predicting the past and future evolutionary space of SARS-CoV-2

Status:

Version 1

Abstract

Figures

Introduction

Methods

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1