Using BUSCO to Assess Insect Genomic Resources

  • Robert M. WaterhouseEmail author
  • Mathieu Seppey
  • Felipe A. Simão
  • Evgeny M. Zdobnov
Part of the Methods in Molecular Biology book series (MIMB, volume 1858)


The increasing affordability of sequencing technologies offers many new and exciting opportunities to address a diverse array of biological questions. This is evidenced in entomological research by numerous genomics and transcriptomics studies that attempt to decipher the often complex relationships among different species or orders and to build “omics” resources to drive advancement of the molecular understanding of insect biology. Being able to gauge the quality of the sequencing data is of critical importance to understanding the potential limitations on the types of questions that these data can be reliably used to address. This chapter details the use of the Benchmarking Universal Single-Copy Orthologue (BUSCO) assessment tool to estimate the completeness of transcriptomes, genome assemblies, and annotated gene sets in terms of their expected gene content.

Key words

Genomics Transcriptomics Genome annotation Completeness assessment Single-copy orthologues 

1 Introduction

Advances in genomics technologies mean that high-throughput nucleotide sequencing has become a relatively low-cost and thus widely accessible tool with numerous applications in biological research. Nevertheless, as researchers in the field know only too well, technical issues, e.g., sample preparation, as well as biological complexities, e.g., large genome sizes, can present substantial challenges to successfully building high-quality genomics resources [1]. Most leading technologies offer in-house sequencing accuracy estimates, and several computational tools allow for detailed assessments of the performance of sequencing and assembly strategies, e.g., QUAST [2] and REAPR [3] genome assembly quality evaluators. Metrics such as contig or scaffold N50 values (half of the total assembly span is made up of contigs or scaffolds of length N50 or longer) offer a summary-statistic view of genome assembly contiguity. Scaffold counts and N50 values from a representative selection of recently published draft insect genomes show that some are currently rather fragmented and will require considerable improvement efforts to reach near-chromosomal-level status (Table 1). However, as a major goal of many genomics studies is to catalogue the complete repertoire of protein-coding genes to facilitate subsequent detailed molecular biology experiments, it is important to also assess the quality of these resources with respect to their completeness in terms of their expected gene content.
Table 1

Assembly statistics and BUSCO assessment results from a representative selection of recently published draft insect genomes



Assembly size (Mbps)

Number of scaffolds

Scaffold N50 (Kbps)

BUSCO completeness


Fruit fly







F:2.5%, M:1.3%

Allen et al.

2017 [4]









F:11.1%, M:7.3%

Davey et al.

2016 [5]


hornworm moth







F:8.4%, M:5.2%

Kanost et al.

2016 [6]









F:0.9%, M:0.8%

Nowell et al.

2017 [7]


fruit fly







F:3.4%, M:1.0%

Papanicolaou et al.

2016 [8]

Bed bug







F:14.0%, M:7.4%

Benoit et al.

2016 [9]

Asian longhorned








F:11.0%, M:3.3%

McKenna et al.

2016 [10]









F:31.8%, M:14.7%

Ioannidis et al.

2017 [11]

BUSCO completeness: C complete, F fragmented, M missing. Species are ordered from the smallest to the largest assembly size and all reported values were retrieved directly from each of the publications listed

The Benchmarking Universal Single-Copy Orthologue (BUSCO ) assessment tool [12, 13] implements such quantifications of completeness for assembled genomes and transcriptomes, as well as annotated protein-coding gene sets. The assessment tool identifies matches to sets of genes that are expected to be present as single-copy orthologues in a given taxonomic group. This expectation is defined by surveying major species clades with numerous sequenced and annotated genomes to identify near-universally-present single-copy orthologues, using the OrthoDB [14] catalogue of orthologues ( For arthropods, BUSCO currently provides five assessment lineages: Arthropoda, Insecta, Endopterygota, Hymenoptera, and Diptera ( The evolutionary filter for genes that are almost always present as single-copy orthologues across a given clade, i.e., genes evolving under “single-copy control” [15, 16], means that they are expected to be present in any newly sequenced species from the same taxonomic group. Quantifying proportions of BUSCOs that can be reliably identified from different genomic resources therefore provides like-for-like estimates of their relative completeness that complement other quality metrics. Importantly, this means that even if a draft genome assembly is still rather fragmented, good BUSCO completeness results allow researchers to proceed with confidence knowing that they have managed to capture most of the expected protein-coding gene repertoire. The examples in Table 1 illustrate how scaffold counts or N50 values are not necessarily predictive of BUSCO completeness, highlighting the importance of such assessments to ensure transparent and intuitive genomic resource quality measures for the benefit of the entire research community.

This chapter presents step-by-step examples of using BUSCO to assess the completeness of different insect genomics resources, with sufficient detail to allow even those unfamiliar with command line computing to run their own assessments. The assessment process consists of running a computational pipeline to identify and then classify BUSCO matches from genome assemblies, annotated gene sets, or transcriptomes, using HMMER [17] hidden Markov models (HMMs). For transcriptomes the longest open reading frames are assessed, while for genome assessments, gene models are first built using ab initio gene prediction with Augustus [18] for the potential matches identified using tBLASTn [19] searches. Matches that meet the BUSCO HMM score cutoffs are classified as “complete” if their lengths fall within BUSCO profile length expectations, and if found more than once they are classified as “duplicated”. Those that do not meet the length requirements are considered as partial matches and are classified as “fragmented”, and BUSCOs without matches that pass the thresholds are classified as “missing”. In this way, the assessments provide an intuitive quantification of the completeness of different genomics datasets in terms of expected gene content.

2 Materials

Before running BUSCO assessments, users are required to first set up the BUSCO software and its dependencies on their computer system and make sure that the data they wish to analyze adhere to the correct formats. These are outlined below, and users are encouraged to visit the website and read the user guide for further detailed information (

2.1 Software Setup

  1. 1.

    BUSCO has been developed in Python and tested on Linux operating systems and it is therefore recommended to use a Linux machine for running BUSCO and its dependencies.

  2. 2.

    The software distribution is available from a public GitLab project where it can be downloaded or preferably (see Note 1 ) cloned using a git client:

    $ git clone

  3. 3.

    As well as Python, the following software packages are BUSCO dependencies and thus must also be installed on the system:

    HMMER (v3.1b2) from
    NCBI BLAST+ from (see Note 2)
    Augustus (v3.2.1 or above) from (see Note 3)

  4. 4.

    BUSCO v3 is installed on the system by executing the script (see Note 4 ):

    $ sudo python install (with root privileges).
    $ python install --user (with only user privileges)

  5. 5.

    BUSCO v3 setup is controlled with a user-editable configuration file. The config.ini.default file in the BUSCO “config” directory must first be copied to config.ini before editing. In this copied file, users must declare the paths to all dependencies (this simply tells BUSCO where they are installed on your system). Users may also use this config.ini file to define the input parameters for a particular analysis, but should be aware that providing input parameters through the command line will override those defined in the config.ini file.

  6. 6.

    Users without access to a Linux machine or cluster may instead use the BUSCO virtual machine (VM). The BUSCO VM was built using OSboxes (, it comes with the BUSCO software and its dependencies already preinstalled and can be downloaded from To run the VM, users need to first download and install a VM manager that is compatible with their system (e.g., Windows, Linux, Macintosh, or Solaris) such as VMware ( or VirtualBox (

  7. 7.

    It is highly recommended to first run a test using the sample data provided as part of the BUSCO software distribution. Execute the following commands and compare the final output “run_TEST” with the provided files in “sample_data/run_SAMPLE”.

    $ python scripts/ --in sample_data/target.fa --out TEST
    --lineage_path sample_data/example --mode genome


2.2 Input Data

  1. 1.

    This chapter details the assessment of several publicly available mosquito genome assemblies and their annotated gene sets from genome sequencing projects [20, 21, 22] that can all be downloaded from VectorBase [23] (

  2. 2.

    The examples used for running transcriptome assessments were selected from publicly available hymenopteran datasets generated as part of large-scale insect transcriptomics studies [24, 25] that can all be downloaded from the NCBI's Transcriptome Shotgun Assembly (TSA) Sequence Database (

  3. 3.

    Input sequence data for genome, transcriptome, or gene set assessments should be provided in standard FASTA format. Files that contain nonstandard nucleotides or amino acids in the sequence lines, or non-alphanumeric or non-ASCII characters in the header lines, could cause errors and therefore these should be avoided wherever possible.

  4. 4.

    Pre-processing of the input data is required in order to obtain true estimates of the numbers of duplicated BUSCOs for annotated gene sets and transcriptomes, these should be pre-processed to select just one representative transcript per gene (see Note 5 ).

  5. 5.

    The lineage datasets used for BUSCO assessments are not provided with the software distribution. Instead, users should download the appropriate lineage dataset(s) from (see Note 6 ). For example, this chapter uses genomic data from dipterans and hymenopterans, so:

    $ wget
    $ wget

  6. 6.

    Each downloaded lineage dataset will need to be unpacked and decompressed before it can be used, for example:

    $ tar -xf diptera_odb9.tar.gz
    $ tar -xf hymenoptera_odb9.tar.gz


3 Methods

With the BUSCO software and its dependencies correctly set up, the relevant BUSCO lineage datasets downloaded and unpacked, and some example insect genomics datasets downloaded and pre-processed (if required), genome, gene set, and transcriptome assessments can now be performed.

3.1 Genome Assessments

  1. 1.

    To run genome mode assessments BUSCO needs to know the location of the Augustus configuration directory, so the “config” path must first be declared as an environment variable (see Note 3 ):

    $ export AUGUSTUS_CONFIG_PATH="/path/to/AUGUSTUS/augustus-3.2.3/config/"

  2. 2.

    The command to launch a genome assembly assessment is made up of four mandatory argument-value pairs that follow the python call to the script:

    --in SEQUENCE_FILE path to your FASTA file, here your genome
    --out NAME a short name that identifies your analysis run
    --lineage_path LINEAGE path to the BUSCO lineage dataset directory
    --mode MODE specify which analysis mode to run, here "genome"

    So to launch an assessment of the genome assembly of the Anopheles arabiensis mosquito using the dipteran lineage dataset the command would be (see Note 7 ):

    $ python /path/to/busco/scripts/
    --in /path/to/mosquito/genomes/Anopheles-arabiensis-D1-genome.fs
    --out AARAD1
    --lineage /path/to/lineage/dataset/diptera_odb9
    --mode genome

  3. 3.

    There are several additional argument-value pairs that are optional and allow users to change the default values of various settings, e.g., if the user’s system has access to multiple computing cores, then they can take advantage of this using the --cpu argument (CPU, central processing unit), or the e-value cutoff for tBLASTn searches can be changed with the --evalue argument (see Note 8 ).

  4. 4.

    One of the most important optional arguments to consider for genome assembly assessments is the choice of Augustus pretrained species-specific gene prediction parameters. Each BUSCO lineage dataset has a predefined default selection, e.g., for the Diptera lineage the default is “fly”, which are Augustus gene prediction parameters pretrained on the fruit fly, Drosophila melanogaster (see Note 9 ).

  5. 5.

    Running the above assessment of the 246.6 megabasepair (Mbp) Anopheles arabiensis genome assembly using the dipteran lineage dataset on 12 CPUs with otherwise default options should take approximately 4 h (see Note 10 ).


3.2 Gene Set Assessments

  1. 1.

    Launching an assessment of an annotated gene set follows the same basic rules as for genome assemblies, with the same four mandatory argument-value pairs that follow the python call to the run_BUSCO .py script:

    --in SEQUENCE_FILE path to your FASTA file, here your proteins
    --out NAME a short name that identifies your analysis run
    --lineage_path LINEAGE path to the BUSCO lineage dataset directory
    --mode MODE specify which analysis mode to run, here "proteins"

    So to launch an assessment of the Anopheles arabiensis annotated gene set (version AaraD1.6) using the dipteran lineage dataset the command would be:

    $ python /path/to/busco/scripts/
    --in /path/to/mosquito/genesets/Anopheles-arabiensis-D1-proteins-1.6.fs
    --out AARAD16
    --lineage /path/to/lineage/dataset/diptera_odb9
    --mode proteins

  2. 2.

    Running this assessment of the 13,452 Anopheles arabiensis (AaraD1.6) protein-coding genes using the dipteran lineage dataset on 4 CPUs with otherwise default options should take approximately 40 min (see Notes 10 and 11 ).


3.3 Transcriptome Assessments

  1. 1.

    Transcriptome assessments are launched with the same four mandatory argument-value pairs that follow the python call to the run_BUSCO .py script:

    --in SEQUENCE_FILE path to your FASTA file, here your transcripts
    --out NAME a short name that identifies your analysis run
    --lineage_path LINEAGE path to the BUSCO lineage dataset directory
    --mode MODE specify which analysis mode to run, here "transcriptome"

    So to launch an assessment of the transcriptome from an adult Pelecinus polyturator parasitoid wasp (NCBI BioProject: PRJNA252202) using the hymenopteran lineage dataset the command would be:

    $ python /path/to/busco/scripts/
    --in /path/to/wasp/transcriptomes/Pelecinus_polyturator.fs
    --out PPOLY
    --lineage /path/to/lineage/dataset/hymenoptera_odb9
    --mode transcriptome

  2. 2.

    Running this assessment of the 35,969 Pelecinus polyturator transcripts using the hymenopteran lineage dataset on 4 CPUs with otherwise default options should take approximately 2 h and 40 min (see Note 10 ). For this analysis the transcriptome was not pre-processed to remove highly similar transcripts (see Note 6 ).


3.4 Understanding the Results

  1. 1.

    Successful assessments will each produce a simple summary results file that reports the full command used to launch the assessment (this is useful in order to be able to re-run the same analysis), as well as the percentages and counts of “complete” (single-copy and duplicated), “fragmented”, and “missingBUSCOs. For example, the results of the Anopheles arabiensis genome assembly assessment:

    2750 Complete BUSCOs (C)
    2746 Complete and single-copy BUSCOs (S)
    4 Complete and duplicated BUSCOs (D)
    19 Fragmented BUSCOs (F)
    30 Missing BUSCOs (M)
    2799 Total BUSCO groups searched

  2. 2.

    All three assessment modes will also produce a “full_table” file with classification results for each BUSCO , a “missing_busco_list” file with the missing BUSCOs, and a “hmmer_output” directory with the full results of the HMM searches. In addition, genome and transcriptome mode assessments will report the results of tBLASTn searches, and genome mode results include the details for all the Augustus gene predictions as well as Augustus training parameters and the nucleotide and protein sequences of the identified complete single-copy BUSCOs (see Note 12 ).

  3. 3.

    The BUSCO plotting tool enables users to visualize their results as a simple bar chart, allowing for clear comparisons of different datasets (see Note 13 ). To generate a chart, users must first copy the short summary results files from each assessment that they want to visualize into a single directory. The script can then be launched pointing to this directory to automatically produce the chart. For example, the commands below will plot the results from the Anopheles arabiensis genome (AARAD1) and gene set (AARAD16) assessments, producing the image file busco_figure.png in the same directory where the summary results were copied:

    $ mkdir arabiensis_results
    $ cp run_AARAD1/short_summary_AARAD1.txt arabiensis_results/.
    $ cp run_AARAD16/short_summary_AARAD16.txt arabiensis_results/.
    $ python /path/to/busco/scripts/ --wd arabiensis_results/

  4. 4.

    Repeating the steps outlined above to assess a total of 15 publicly available mosquito genome assemblies and their annotated gene sets and then plotting the results enables the like-for-like comparison of these genomic resources, where all but five datasets are more than 95% “complete” (Fig. 1). Furthermore, these mosquito genomics resources all show very low levels of duplications, indicating that the assemblies are likely mostly free of haplotype regions (see Note 14 ). In addition, the genome assembly results generally mirror those of the gene sets, with the assemblies usually performing slightly better apart from a few cases where the assembly appears substantially better than the gene set (see Note 15 ). These estimates of “complete”, “fragmented”, and “missingBUSCOs (see Note 16 ) provide intuitive metrics with which to gauge the relative quality of these genomic resources in terms of their expected gene content.

  5. 5.

    Repeating the steps outlined above for the assessments of many more publicly available hymenopteran transcriptomes and plotting the resulting completeness scores against the numbers of transcripts demonstrates their highly variable completeness (Fig. 2). Transcriptomes may well be expected to show rather variable completeness scores as the total repertoire of RNAs that are sequenced and assembled will often reflect the type of biological sample, e.g., a pooled sample from multiple tissues and life-stages will probably capture more than a sample from a specialized tissue (see Note 17 ).

Fig. 1

BUSCO results from the assessments of 15 Anopheles mosquito genomes and their annotated gene sets. The chart was produced using the BUSCO plotting tool and demonstrates the intuitive visualization of side-by-side genome and gene set results from multiple species. Gene set results (e.g., AARAD16) are plotted directly beneath genome assembly results (e.g., AARAD1) for each species with numbers indicating dataset versions: An. arabiensis (AARAD), An. atroparvus (AATRE), An. coluzzii (ACOLM), An. culicifacies (ACULA), An. dirus (ADIRW), An. epiroticus (AEPIE), An. farauti (AFARF), An. funestus (AFUNF), An. gambiae (AGAMP), An. merus (AMERM), An. minimus (AMINM), An. quadriannulatus (AQUAS), An. sinensis (ASINS), An. stephensi Indian (ASTEI), An. stephensi SDA-500 (ASTES)

Fig. 2

BUSCO completeness results from the assessments of 103 hymenopteran transcriptomes from Peters et al. 2017 [24] and Petersen et al. 2017 [25] compared to the number of transcripts in each transcriptome. Transcriptomes with many transcripts are not necessarily the most complete, and those with fewer transcripts can still score relatively well in terms of completeness. Several example species are labeled either directly to the left or right of the data point or centered directly above or below it

4 Notes

  1. 1.

    Users are encouraged to use the git client option to retrieve the BUSCO software as this will make installation of future updates much simpler and easy to manage. Additionally, the GitLab project “issues” page is worth consulting as it is a good source of tips and discussions from BUSCO users.

  2. 2.

    It has been reported that when running BUSCO using multiple cores, the tBLASTn step from BLAST+ versions 2.4, 2.5, and 2.6 may occasionally fail to complete and thus the BUSCO assessment will fail with an error message to this effect. To avoid this problem, use an earlier BLAST+ version or run using only a single core.

  3. 3.

    Users only need to install Augustus if they plan to assess genome assemblies. As Augustus has dependencies of its own, e.g., Perl, users should consult the Augustus documentation for the correct installation procedures. If working on a system where Augustus has already been installed by an administrator and the user does not have “write permission” to the Augustusconfig” directory, users can simply recursively copy the entire “config” directory to a location where they do have “write permission” and then re-set the “config” path variable to this location:

    $ cp -r /path/to/AUGUSTUS/augustus-3.2.3/config /my/home/augustus/config
    $ export AUGUSTUS_CONFIG_PATH="/my/home/augustus/config/"

  4. 4.

    This was not a requirement for BUSCO v1 or v2. The v3 update refactored the underlying analysis code to make it more modular and extendable and thus it must be installed using the script.

  5. 5.

    For annotated gene sets the transcript-to-gene relationships are defined in the annotation files, e.g., General Feature Format (GFF) files, so the longest protein-coding transcript can be selected for each gene with multiple annotated transcripts. For de novo transcriptomes, i.e., those without a reference genome, transcript-to-gene relationships are not defined so users have two options: (a) run the assessments without pre-processing and acknowledge the fact that estimates of duplicated BUSCOs are likely to be inflated by the presence of multiple transcripts from the same gene, or (b) pre-process the transcriptome with a sequence identity (or similarity) and length filter to select just one representative from sets of highly similar transcripts, e.g., using CD-Hit [26].

  6. 6.

    There are currently 16 bacterial lineage datasets and 28 eukaryotic lineage datasets. Users would normally select the most specific lineage available, i.e., the most recent ancestor of the species whose data is to be assessed. For example, for assessing ant data one would select the “hymenoptera” lineage rather than the “arthropoda” lineage. However, if there are a large number of species/strains/versions, etc. to be assessed, then to minimize runtime (at the expense of resolution) one might select a less specific (more ancestral) lineage dataset with fewer BUSCOs, at least for the initial rounds of assessments.

  7. 7.

    BUSCO outputs the running log details to the default standard output (user’s terminal), in order to instead send these details to a file users can simply end the launch command with a redirect command, and as BUSCO assessments can take some time it is useful to run them in the background, i.e., end launch command with: >& my_log_file.txt &.

  8. 8.

    The optional arguments for launching a BUSCO assessment give the user flexibility over many aspects, some specific to running in genome mode and others applicable in any mode, all of which are described in full in the user guide. Some useful options to consider employing include (a) --force, this will force the results to overwrite results from an analysis run with the same name (b) --tarzip, this will package and compress the results from steps that can produce many output files; (c) --augustus_options, this allows users to pass Augustus-specific parameters for gene prediction, e.g., to use alternative codon translation tables.

  9. 9.

    Augustus comes with pretrained gene prediction parameters for many species (see Augustus documentation for up-to-date details), so if parameter sets are available for the species to be assessed, then they should be selected, e.g., for the Florida carpenter ant, the parameter set to use can be specified by adding the argument “--species camponotus_floridanus” to the launch command. For many other species, pretrained gene prediction parameters are not yet available, so users should select the closest species for which such parameters are available, or run the assessment with the preselected default parameters. For the sake of reproducibility it is important to specify which one was selected when reporting BUSCO results.

  10. 10.

    Assessment runtimes will vary according to the exact system setup. Assessments of genome assemblies require the initial steps of first identifying genomic regions that potentially harbor BUSCO matches and then predicting gene models in these regions. These are computationally intensive tasks and therefore genome assembly assessments will take substantially longer than transcriptome or gene set assessments. Note also that the searches and gene predictions are performed in two rounds: (a) searches with consensus sequences built from BUSCO HMMs followed by gene predictions using the selected Augustus pretrained parameter set; (b) then for BUSCOs that were classified as “fragmented” or “missing” after the first round, searches with variant consensus sequences followed by gene predictions using parameters trained on “completeBUSCOs identified in round one. Thus, if the first round identifies a high proportion of “completeBUSCOs, then the second round will be relatively quick, but if there are many “fragmented” or “missingBUSCOs after the first round, then the second round will take considerably longer.

  11. 11.

    The Anopheles arabiensis AaraD1.6 annotation contains 13,452 protein-coding genes with 13,640 transcripts so the protein FASTA file downloaded from VectorBase was first pre-processed to select the longest protein per gene. Performing this pre-processing step on annotated gene sets is not obligatory, but it ensures that BUSCO estimates of the numbers of duplicated genes will be true assessments that are not inflated by alternative transcripts that would be reported as gene duplicates.

  12. 12.

    During genome assessments the second round of gene predictions uses parameter sets built from the “completeBUSCOs identified in the first round. These Augustus retraining parameters are saved in the “augustus_output” results directory. They are ideal for use during whole genome annotation procedures that employ Augustus, especially when parameter sets for the species to be annotated or those of a close relative are not already available. In addition, the GenBank or GFF formatted “completeBUSCO annotations provided in the results directory can be used to train other gene predictors, e.g., SNAP [27].

  13. 13.

    The BUSCO plotting tool uses R ( and the ggplot2 library (, so these must be installed and accessible on the system in order to produce the image. Alternatively, adding the optional argument --no_r to the command will simply produce the R script required to build the image and users can then run this R script on any system where R and ggplot2 are installed. This also gives the user the opportunity to edit the R script to tailor the resulting image, e.g., changing the default fonts, labels, or colors.

  14. 14.

    If high levels of complete duplicates are reported for a genome assembly, then this could suggest that the assembly procedure has failed to correctly collapse haplotype regions, resulting in numerous pairs of highly similar duplicate gene copies. This would warrant further investigations to determine if this is indeed the case and if alternative assembly strategies need to be employed or if such regions can be removed or collapsed. However, knowledge of the biology of the sample itself can also offer explanations: e.g., assessing the Aedes albopictus C6/36 cell line genome assembly and annotated gene set suggested that most BUSCOs were duplicated, but cytogenetic studies have shown that this cell line does have aberrant karyotypes, which could explain the numerous duplicates [28].

  15. 15.

    Differences in the results from assessing a genome assembly versus its annotated gene set may be due to several factors. In both cases BUSCO attempts to classify the matches to a set of protein-coding gene annotations: for genomes these annotations are built by BUSCO using Augustus gene predictions with BUSCO HMMs, whereas for gene sets they have usually been built by genome annotation pipelines (e.g., MAKER [29]) that incorporate evidence from several gene predictors and different sources of gene model support. Thus when results from assembly assessments appear to be better than for their gene sets it suggests that the targeted approach taken by BUSCO has produced better gene models than a more generalist annotation pipeline (at least for the subset of genes that make up the BUSCO lineage dataset). Conversely, if a gene set appears more complete than its genome, this suggests that the multiple sources of evidence used by the annotation pipeline have resulted in generally better annotations than the single-predictor approach taken by BUSCO.

  16. 16.

    When interpreting BUSCO results, users should be aware that the classification procedure (described in the introduction) results in the labels “complete”, “fragmented”, and “missing”, which are by necessity simplifications that reflect the most likely scenario. For example, the label “missing” is applied to BUSCOs with no matches (probably truly absent from the dataset), but also to matches that do not meet the HMM score cutoffs. These below-cutoff matches could mean that these BUSCOs are in fact partially present in the dataset but there is simply not enough matching sequence be confident of the partial match and classify them as “fragmented”.

  17. 17.

    BUSCO assessments are usually performed to demonstrate the good completeness levels of the genomic resources generated and analyzed in a particular study. However, if the aim of a transcriptomics experiment is to sample a specific tissue or life-stage where the repertoire of transcripts is expected to be highly specialized, then low completeness scores would in fact offer support that such targeted sampling was successful.




R.M.W. was supported by Swiss National Science Foundation award PP00P3_170664.


  1. 1.
    Richards S, Murali SC (2015) Best practices in insect genome sequencing: what works and what doesn’t. Curr Opin Insect Sci 7:1–7CrossRefGoogle Scholar
  2. 2.
    Gurevich A, Saveliev V, Vyahhi N et al (2013) QUAST: quality assessment tool for genome assemblies. Bioinformatics 29:1072–1075CrossRefGoogle Scholar
  3. 3.
    Hunt M, Kikuchi T, Sanders M et al (2013) REAPR: a universal tool for genome assembly evaluation. Genome Biol 14:R47CrossRefGoogle Scholar
  4. 4.
    Allen SL, Delaney EK, Kopp A et al (2017) Single-Molecule Sequencing of the Drosophila serrata Genome. G3: Genes, Genomes, Genetics 7:781–788CrossRefGoogle Scholar
  5. 5.
    Davey JW, Chouteau M, Barker SL et al (2016) Major improvements to the Heliconius melpomene genome assembly used to confirm 10 chromosome fusion events in 6 million years of butterfly evolution. G3: Genes, Genomes, Genetics 6:695–708CrossRefGoogle Scholar
  6. 6.
    Kanost MR, Arrese EL, Cao X et al (2016) Multifaceted biological insights from a draft genome sequence of the tobacco hornworm moth, Manduca sexta. Insect Biochem Mol Biol 76:118–147CrossRefGoogle Scholar
  7. 7.
    Nowell RW, Elsworth B, Oostra V et al (2017) A high-coverage draft genome of the mycalesine butterfly Bicyclus anynana. GigaScience 6:1–7CrossRefGoogle Scholar
  8. 8.
    Papanicolaou A, Schetelig MF, Arensburger P et al (2016) The whole genome sequence of the Mediterranean fruit fly, Ceratitis capitata (Wiedemann), reveals insights into the biology and adaptive evolution of a highly invasive pest species. Genome Biol 17:192CrossRefGoogle Scholar
  9. 9.
    Benoit JB, Adelman ZN, Reinhardt K et al (2016) Unique features of a global human ectoparasite identified through sequencing of the bed bug genome. Nat Commun 7:10165CrossRefGoogle Scholar
  10. 10.
    McKenna DD, Scully ED, Pauchet Y et al (2016) Genome of the Asian longhorned beetle (Anoplophora glabripennis), a globally significant invasive species, reveals key functional and evolutionary innovations at the beetle-plant interface. Genome Biol 17:227CrossRefGoogle Scholar
  11. 11.
    Ioannidis P, Simão FA, Waterhouse RM et al (2017) Genomic features of the damselfly Calopteryx splendens representing a sister clade to most insect orders. Genome Biol Evol 9:415–430PubMedPubMedCentralGoogle Scholar
  12. 12.
    Simão FA, Waterhouse RM, Ioannidis P et al (2015) BUSCO: Assessing genome assembly and annotation completeness with single-copy orthologs. Bioinformatics 31:3210–3212CrossRefGoogle Scholar
  13. 13.
    Waterhouse RM, Seppey M, Simão FA et al (2017) BUSCO applications from quality assessments to gene prediction and phylogenomics. Mol Biol Evol 35:543–548Google Scholar
  14. 14.
    Zdobnov EM, Tegenfeldt F, Kuznetsov D et al (2017) OrthoDB v9.1: Cataloging evolutionary and functional annotations for animal, fungal, plant, archaeal, bacterial and viral orthologs. Nucleic Acids Res 45:D744–D749CrossRefGoogle Scholar
  15. 15.
    Waterhouse RM, Zdobnov EM, Kriventseva EV (2011) Correlating traits of gene retention, sequence divergence, duplicability and essentiality in vertebrates, arthropods, and fungi. Genome Biol Evol 3:75–86CrossRefGoogle Scholar
  16. 16.
    Waterhouse RM (2015) A maturing understanding of the composition of the insect gene repertoire. Curr Opin Insect Sci 7:15–23CrossRefGoogle Scholar
  17. 17.
    Eddy SR (2011) Accelerated Profile HMM Searches. PLoS Comput Biol 7:e1002195CrossRefGoogle Scholar
  18. 18.
    Keller O, Kollmar M, Stanke M et al (2011) A novel hybrid gene prediction method employing protein multiple sequence alignments. Bioinformatics 27:757–763CrossRefGoogle Scholar
  19. 19.
    Camacho C, Coulouris G, Avagyan V et al (2009) BLAST+: architecture and applications. BMC Bioinformatics 10:421CrossRefGoogle Scholar
  20. 20.
    Holt RA, Subramanian GM, Halpern A et al (2002) The genome sequence of the malaria mosquito Anopheles gambiae. Science 298:129–149CrossRefGoogle Scholar
  21. 21.
    Jiang X, Peery A, Hall AB et al (2014) Genome analysis of a major urban malaria vector mosquito, Anopheles stephensi. Genome Biol 15:459CrossRefGoogle Scholar
  22. 22.
    Neafsey DE, Waterhouse RM, Abai MR et al (2015) Highly evolvable malaria vectors: the genomes of 16 Anopheles mosquitoes. Science 347:1258522–1258522CrossRefGoogle Scholar
  23. 23.
    Giraldo-Calderón GI, Emrich SJ, MacCallum RM et al (2015) VectorBase: an updated bioinformatics resource for invertebrate vectors and other organisms related with human diseases. Nucleic Acids Res 43:D707–D713CrossRefGoogle Scholar
  24. 24.
    Peters RS, Krogmann L, Mayer C et al (2017) Evolutionary history of the hymenoptera. Curr Biol 27:1013–1018CrossRefGoogle Scholar
  25. 25.
    Petersen M, Meusemann K, Donath A et al (2017) Orthograph: a versatile tool for mapping coding nucleotide sequences to clusters of orthologous genes. BMC Bioinformatics 18:111CrossRefGoogle Scholar
  26. 26.
    Li W, Godzik A (2006) CD-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences. Bioinformatics 22:1658–1659CrossRefGoogle Scholar
  27. 27.
    Korf I (2004) Gene finding in novel genomes. BMC Bioinformatics 5:59CrossRefGoogle Scholar
  28. 28.
    Waterhouse RM, Chen X, Bonizzoni M et al (2017) The third International Workshop on Aedes albopictus: building scientific alliances in the fight against the globally invasive Asian tiger mosquito. Pathog Global Health 111:161–165CrossRefGoogle Scholar
  29. 29.
    Campbell MS, Holt C, Moore B et al (2014) Genome annotation and curation using MAKER and MAKER-P. Curr Protoc Bioinformatics 48:4.11.1-39PubMedGoogle Scholar

Copyright information

© Springer Science+Business Media, LLC, part of Springer Nature 2019

Authors and Affiliations

  • Robert M. Waterhouse
    • 1
    Email author
  • Mathieu Seppey
    • 2
  • Felipe A. Simão
    • 2
  • Evgeny M. Zdobnov
    • 2
  1. 1.Department of Ecology and EvolutionUniversity of Lausanne, and Swiss Institute of BioinformaticsLausanneSwitzerland
  2. 2.University of Geneva and Swiss Institute of BioinformaticsGenevaSwitzerland

Personalised recommendations