Biobtree: A tool to search, map and visualize bioinformatics identifiers and special keywords.

Due to their nature, bioinformatics datasets are often closely related to each other. For this reason, search, mapping and visualization of these relations are often performed manually or programmatically via identifiers or special keywords such as gene symbols. Although various tools exist for these situations, the growing volume of bioinformatics datasets, emerging new software tools and approaches motivates new solutions. To provide a new tool for these current cases, I present the Biobtree bioinformatics tool. Biobtree effectively fetches and indexes identifiers and special keywords with their related identifiers from supported datasets, optionally with user pre-defined datasets and provides a web interface, web services and direct B+ tree data structure based single uniform database output. Biobtree can handle billions of identifiers and runs via a single executable file with no installation and dependency required. It also aims to provide a relatively small codebase for easy maintenance, addition of new features and extension to larger datasets. Biobtree is available to download from GitHub. Copyright:

Introduction

Mapping bioinformatics datasets through a web interface or programmatically via identifiers or special keywords and attributes such as gene name, gene location, protein accessions and species name is a common need during genomics research. These mappings play an essential role in molecular data integration ( Huang ) and allow the gathering of maximum biological insight ( Mudunuri ) for these diverse bioinformatics datasets.

There are several existing tools for these mapping needs; these tools are gene-centric, protein-centric or can provide both gene- and protein-centric solutions. One of the common gene-centric tools is BioMart ( Zhang )-based tools such as Ensembl BioMarts ( Kinsella ) which covers Ensembl ( Zerbino ) and Ensembl Genomes ( Kersey ) datasets. The R programming language package biomarRt ( Durinck ) is also widely used via performing queries with BioMart-based tools. Other common gene-centric tools are MyGene.info ( Xin ), DAVID ( Huang da ) and g:Profiler ( Raudvere ). Uniprot ID mapping service ( Huang ) provides a protein-centric solution. bioDBnet ( Mudunuri ) and BridgeDb ( van Iersel ) provide services for both gene- and protein-centric solutions.

On the other hand, genomics data size is increasing continuously ( Langmead & Nellore, 2018) especially via high throughput sequencing, so performing these mappings on these expanding data sizes in local computers, cloud computing or existing computing environments in a rapid and effective way via tools with easy installation and requiring minimum maintenance is a challenge ( Marx, 2013).

The referenced existing gene-centric tools currently do not support large Ensembl Bacteria genomes. Existing tools either provide only online services or require specific technical knowledge such as a particular database or specific programming language to install, use and adapt to different computational environments such as a local computer. Another limitation of the referenced tools is that they provide one-dimensional filtering capability in a single mapping query.

Biobtree addresses these problems of existing tools, First, it can be used via a single executable file without requiring re-compilation or extra maintenance such as database administration. Alternatively, it can be used via the R or Python-based wrapper packages which have been provided to allow for easier integration into existing pipelines. To process large datasets, it uses a specialized MapReduce-based solution which is discussed in the next section. MapReduce is an effective way to deal with large datasets ( Langmead & Nellore, 2018). After processing data, Biobtree provides a web interface, web services and chain mapping and filtering query capability in a single query with its intuitive query syntax which is demonstrated in the use cases section. Biobtree covers a range of bioinformatics datasets including Ensembl Bacteria genomes. The data resources currently used are ChEBI ( Hastings ), HGNC ( Braschi ), HMDB ( Wishart ), InterPro ( Mitchell ), Europe PMC ( Europe PMC Consortium, 2015), UniProt ( UniProt Consortium, 2019), Chembl ( Gaulton ), Gene Ontology ( The Gene Ontology Consortium, 2019), EFO ( Malone ), ECO ( Giglio ), Ensembl ( Zerbino ) and Ensembl Genomes ( Kersey ). Table 1 shows details of these datasets.

Table 1.

List of datasets.

Dataset	Description	Location	Format
ChEBI	ChEBI reference accession data	ftp.ebi.ac.uk/chebi/Flat_file_tab_delimited/	TSV
HGNC	Human gene nomenclature	ftp.ebi.ac.uk/genenames/new/json/	JSON
HMDB	Human metabolome database	http://www.hmdb.ca/system/downloads/current/	XML
InterPro	Protein Families	ftp://ftp.ebi.ac.uk/pub/databases/interpro/current	XML
Literature mappings	Literature pmid, pmcid and doi mappings	ftp://ftp.ebi.ac.uk/pub/databases/pmc/DOI/	CSV
Taxonomy	NCBI Taxonomy	ftp://ftp.ebi.ac.uk/pub/databases/taxonomy/	XML
Uniparc	UniProt Sequence Archive	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/uniparc/	XML
UniProt reviewed	UniProt Knowledgebase reviewed	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/knowledgebase/complete/	XML
UniProt unreviewed	UniProt Knowledgebase unreviewed	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/knowledgebase/complete/	XML
Uniref50	UniProt sequence clusters	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/uniref/uniref50/	XML
Uniref90	UniProt sequence clusters	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/uniref/uniref90/	XML
Uniref100	UniProt sequence clusters	ftp.ebi.ac.uk/pub/databases/uniprot/current_release/uniref/uniref100/	XML
GO	Gene Ontology	http://purl.obolibrary.org/obo/go.owl	RDF/XML
ECO	The Evidence & Conclusion Ontology	http://purl.obolibrary.org/obo/eco.owl	RDF/XML
EFO	Experimental Factor Ontology	http://www.ebi.ac.uk/efo/efo.owl	RDF/XML
ChEMBL	Chemical database of bioactive molecules	ftp.ebi.ac.uk/pub/databases/chembl/ChEMBL-RDF/latest/	RDF/XML
Ensembl	Ensembl	ftp.ensembl.org/pub/current_json/ ftp.ensembl.org/pub/current_mysql/ ftp.ensembl.org/pub/current_gff3/	JSON,CSV, GFF3
Ensembl Genomes Metazoa	Ensembl Genomes Metazoa	ftp://ftp.ensemblgenomes.org/pub/current/metazoa/json/ ftp://ftp.ensemblgenomes.org/pub/current/metazoa/mysql/ ftp://ftp.ensemblgenomes.org/pub/current/metazoa/gff3/	JSON,CSV, GFF3
Ensembl Genomes Plants	Ensembl Genomes Plants	ftp://ftp.ensemblgenomes.org/pub/current/plants/json/ ftp://ftp.ensemblgenomes.org/pub/current/plants/mysql/ ftp://ftp.ensemblgenomes.org/pub/current/plants/gff3/	JSON,CSV, GFF3
Ensembl Genomes Fungi	Ensembl Genomes Fungi	ftp://ftp.ensemblgenomes.org/pub/current/fungi/json/ ftp://ftp.ensemblgenomes.org/pub/current/fungi/mysql/ ftp://ftp.ensemblgenomes.org/pub/current/fungi/gff3/	JSON,CSV, GFF3
Ensembl Genomes Protists	Ensembl Genomes Protists	ftp://ftp.ensemblgenomes.org/pub/current/protists/json/ ftp://ftp.ensemblgenomes.org/pub/current/protists/mysql/ ftp://ftp.ensemblgenomes.org/pub/current/protists/gff3/	JSON,CSV, GFF3
Ensembl Genomes Bacteria	Ensembl Genomes Bacteria	ftp://ftp.ensemblgenomes.org/pub/current/bacteria/json/ ftp://ftp.ensemblgenomes.org/pub/current/bacteria/gff3/	JSON,GFF3

Methods

Implementation

The Biobtree implementation process starts by retrieving selected datasets as shown in Table 1 and retrieving data entries belonging to these datasets with their attributes and mapping information from their public locations, which are also shown in Table 1. During this data retrieval, the whole of the data do not get stored and uncompressed on the disk, instead data are retrieved and uncompressed in a streaming manner in the memory, which allows avoiding excessive disk space usage. Necessary data, which are these mapping and attributes, are compactly stored as chunks on the disk. During these data retrievals, all the idle CPU cores have been utilized to merge and sort these chunks recursively with each other. It is essential that the produced files are sorted to make fast batch inserts to the LMDB database which Biobtree uses as a database to store its result data. Once the data retrieval process is completed, result chunk files are globally merged using the patience sort technique and inserted into the LMDB database as keys and values. Keys consist of identifiers and special keywords such as gene names or species name, and values are attributes such as genomic coordinates and mapped identifiers. In these processes, data retrieval and creation of sorted chunks represent the map phase, global merge of the chunks and database creation represent the reduce phase of the MapReduce solution. Once the database is created, the Biobtree web module provides a web interface and web services to perform both searching for identifiers and mapping queries. Mapping queries has been done with a query syntax which allows chains of mapping and filtering between datasets. An example use case with this syntax is demonstrated in the next section. Biobtree uses a B+ tree data structure-based LMDB key-value store. LMDB provides fast batch inserts and reads which fits the bioinformatics datasets update cycle well where datasets are often updated periodically, and then only intensive read operations are performed. LMDB is embedded into Biobtree’s executable binary code so it does not require a separate installation or special maintenance.

Use cases

The Biobtree web interface can be used primarily for exploration purposes and web services related to integrating genomic analysis pipelines via the Biobtree executable file which is available at GitHub. However, use of the Biobtree executable requires some learning of Biobtree usage and also, in relation to integrating into pipelines written in the various different languages, requires some extra coding effort.

In order to address the above situations, R and Python based wrapper packages BiobtreeR and BiobtreePy have been provided. R and Python are commonly used in genomic analysis ( Russell ). Both the BiobtreeR and BiobtreePy packages provide very similar functionalities and allow Biobtree to be used seamlessly within pipelines written in these languages.

The BiobtreeR package is provided via Bioconductor ( Huber ) and meets the build, test and quality standards stipulated for a Bioconductor package.

Another usability feature aimed at easier integration with existing pipelines, which was suggested in the course of the BiobtreeR package review process and has been implemented, is that of providing built-in Biobtree databases for commonly studied datasets and organism genomes. This feature is intended to speed up the data build and update processes related to common datasets and organism genomes and includes example use cases via the web interface. These latter serve the purpose of familiarizing users with the Biobtree and its query syntax.

The following three use cases are demonstrated using BiobtreeR and involve its installation instructions. The first two use cases employ built-in databases, but in the last use case data is built from genomes belonging to specific taxonomy identifiers which are not included in the built-in databases.

BiobtreeR installation

usecase-1 Map Affymetrix identifiers to Ensembl human genome identifiers and then map these to the molecular function type GO term s

usecase-2 Map human Ensembl identifiers with given genome location to the reviewed Uniprot identifiers

usecase-3 Map all taxonomic children of given bacteria and then map these children to Ensembl with given genome location and contains a given word

Discussion

A mapping between bioinformatics datasets via identifiers or special keywords such as species names is often performed during genomic analyses and plays an essential role in molecular data integration and getting maximum biological insight from these datasets. There are several gene-centric, protein-centric and both protein- and gene-centric tools for addressing these mapping needs. These tools currently do not support the large Ensembl Genomes Bacteria dataset. In addition, these tools provide either only online services or require specific technical knowledge to install and adapt to new computing environments. Existing tools also provide one-dimensional filtering in a single mapping query. Biobtree addresses these problems by managing a tool with a single executable file or alternatively additionally provided R or Python based wrapper packages and processing large datasets with its specialized MapReduce-based solution. Based on processed data, it creates a uniform database and allows searching identifiers and chain mappings and filtering queries.

Data availability

All data underlying the results are available as part of the article and no additional source data are required.

Software availability

All source codes and binaries available at: https://www.github.com/tamerh/biobtree.

Archived source code at time of publication: https://doi.org/10.5281/zenodo.2547047

License: BSD 3-Clause “New” or “Revised” license.

I hereby again confirm my approval status. Great work!

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

The author has made additional efforts to improve the installation and use of the tool. I am happy to accept it and hope the author will continue to improve and maintain it over time.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

Thank you for your review and suggestions to make the article accepted. I have submitted a new version of the article for the English language issues. And yes I will maintain and improve the tool. Especially maintenance is necessary for the Bioconductor package. I will also add major new features if I get requests from users.

I hereby confirm my approval status of this article, while suggesting to fix the following language issues:

Introduction:

"performing these mapping" --> "performing these mappings".

"provide only online service" --> "provide only online services".

"all the idle CPUs" --> "all the idle CPU cores".

(This is also to be more technically correct, as few computers these days have multiple CPUs, while most of them have CPUs with multiple cores).

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

Thank you again for your review and suggestions to make the article accepted.

I have submitted the new version for the English language issues.

The revised article is now much improved, with a solid introduction explaining the problem area and previous research, use cases, and with the title amended to better reflect the functionality of the tool. I think it can now be approved, with the following minor suggestions for language fixes:

Abstract

web interface -> a web interface

Introduction

P. 3: Biobtree address -> Biobtree addresses

Methods

I think "disc" should be "disk" for hard drive disks.

as key and values -> as keys and values

"... and values are attributes and mapped datasets information".

This is a somewhat unclear sentence. I.e, does it mean "mapped datasets' information" (genitive), or "information about mapped datasets"? I suggest slightly clarifying it.

Use cases

"... third input is dataset to" -> "... third input is a dataset to"

"in different dataset" -> "different datasets" or "a different dataset"

"with same value" -> "with the same value"

Future work

"In addition, following and experimenting with the advancements in large data processing techniques, databases and data structures fields to improve the tool further".

This sentence seems incomplete. I.e, what is stated about this activity? Something that is planned to be done, or "could be done in general"?

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.

Thank you very much for taking time to review the article and for all the comments and suggestions. I will address your new minor suggestions in the next version together with demonstrating the use cases with R package.

The author now includes references to previously published annotation tools and services such as BioMart and MyGene.info and points out several limitations of the extant tools that are addressed with Biobtree. In addition, the author now includes an online portal for exploring data with a structured query language along with a few examples of specific use cases and a table of databases implemented.

Given the improvements, the work can now be beneficial for biologists interested in mapping and annotation of specific genes or terms, which can serve as an alternative to already established annotation databases and services.

However, there are still several limitations to this work. First, supposedly one of the main advantages of Biobtree is that it can be run as a standalone executable tool. This is great in theory but there doesn't seem to be any executables in the referenced repository or an explanation of how to install the Biobtree in various environments. The manuscript goes into some detail about the underlying algorithms and datastructures, but there is no explanation of what is required for you to install the program or get it working in either the manuscript or the repository. Therefore, without documentation or description the only way to use the tool currently is through the online portal. The second major limitation is that while there are three use-cases described, there is no documentation provided for how one would build arbitrary queries. A manual or easier to use interface is needed before the full power of the tool can be leveraged. Finally, I couldn't find a way to download the results from the online portal. There is often hundreds of pages of results but it seems there is no way to obtain a table of the results for use in downstream tools or analyses. Given the fact that there is no description of how to install the tool or which prerequisites are needed, and the fact that you can't download the results from the online portal, this tool is currently only useful for manual exploration of the mappings and not suitable for inclusion in bioinformatics pipelines. Since it seems the author intended for the latter, instructions for how to install the tool, a detailed description of the API or commands, and a way to save the output to a table of some kind is still needed. I look forward to seeing the revised manuscript.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

The article describes a commandline tool, Biobtree, that is claimed to allow to process relations between bioinformatics datasets based on various characteristics such as identifiers and keywords.

The manuscript describes the tool in a clear way technically, making it quite clear what it does in technical terms, and how it is supposed to be used.

Also, I was able to install and run the tool in a simple way on my laptop (i5 CPU, 8GB RAM and 10-15 GB free hard drive, Xubuntu 16.04 64 bit) without problems. It provides a simple but good looking and easy to use web interface.

I'm seeing at least two major issues with the tool and manuscript though, that needs being thoroughly addressed to make them acceptable.

Main problem 1: Visualization?

Firstly, the title claims that the tool does visualization of the database produced by the tool. Perhaps I'm missing something, but I have not found any visualization in the tool apart from a form of search hit result listings. I don't think this is enough to be called "visualization". Especially as it is unclear how the current form of output is supposed to be used in a concrete biological usecase. With the current wording, I would expect something more graphical, like a graphviz-like graph view of dataset relations.

Suggested edits to make the tool and paper acceptable: Main problem 2: Lack of context and discussion of biological relevance

Provide graphical visualization beyond results listings (or explain how to show them, if I have missed them), or else remove "visualization" from the title and other places.

Use this/these visualizations in the use cases/demonstrators discussed above, to explain how they contribute to solving concrete biological problems.

The first and main problem with the manuscript is that it does not provide a clear enough description of what biological problem it is solving. Nor does it provide an overview of existing tools and solutions in this field. Right now, the manuscript only states what the tool can do in technical terms. It somehow reads like a (well written) user guide or README file, but not yet a scientific paper. To help potential new users understand why they might need this tool, it needs to be put in context and compared with other existing tools.

In my view, the manuscript needs the following points thoroughly addressing to be acceptable: Language issues

In the introduction: Elaborate on the field of mapping/visualising dataset relations, mentioning relevant existing similar tools, what are the typical problems, and what particular problem Biobtree solves.

Explain a few examples of biological problems that can be solved with this tool, or type of tool.

E.g. in the results: Provide at least one, and optimally two or three potentially simple, but relevant, biological demonstrators or use cases, that can be addressed with the tool. Provide complete instructions on how to re-run this or these demo(s) and provide outputs for this/these in terms of figures or diagrams and how these were produced. In this way, both reviewers and users can make sure that they understand how to operate the tool.

In the discussion: Connect back to the explained problem the tool is addressing, and explain how the problem was solved, again reinstating the relevance of this specific tool compared to other existing tools, and what improvement it provides to the end user trying to solve biological problems, exemplified by the demonstrators or use cases.

The manuscript also contains quite a number of language issues. I'm listing a few language suggestions below as examples, but further language proofing or editing is highly recommended, to make sure there are not more of these:

Methods section:

"in GO programming language" --> "in the Go programming language"

(Note the "the" and that only G is uppercase in "Go").

Update phase section:

"to LMDB" -> "to the LMDB"

Update phase section:

"Updating reads selected datasets as a stream"

I don't understand this sentence. Please language-check it.

"is used in next" -> "is used in the next"

Generate phase section:

"the project github page" -> "the project's GitHub page"

Web phase section:

"starts web phase" -> "starts the web phase"

Web interface section:

"The Web interface allow user" -> "The web interface allows the user"

Operation section:

"Biobtree executable" -> "The biobtree executable".

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.

While it is important to create a consistent and queryable database of biological identifiers, it is unclear what advances this tool brings to the field. For example, how does this tool compare to other queryable database tools such as mygene.info, or BioMart? The paper will greatly benefit from a comparison to these and other such tools.

I downloaded and ran the tool but it seems I can't get through the update phase when I run ./biobtree update (It seems to hang after uniprot_reviewed finishes) without any other messages. When I rerun using biobtree --d uniprot_reviewed update it finishes but there is an error:

Error while reading file-> .//out/index/0_13.938476000.gz

panic: gzip: invalid header

I tried running generate and web after that regardless, but couldn't get it to work:

panic: mdb_txn_commit: MDB_BAD_TXN: Transaction must abort, has a child, or is invalid

Error while reading meta information file which should be produced with generate command. Please make sure you did previous steps correctly.

The author needs to debug/test their code to ensure that it can be used by others.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above.

Thank you for reviewing the article. I agree that there are several similar tools exist with different dataset and functionalities such as Biomart and mygene.info. However, this tool can still complement them for following main reasons. In relation to reported errors, I have added demo of tool in case such errors happen again. I have also added integration test which runs periodically on Linux, MacOS and Windows operating systems via Azure DevOps platform. These tests can be accessed publicly and test and demo links can be found at github page. Based on these tests, it seems that tool is working as expected. I believe that hanged process is a specific issue or bug which I am happy to resolve if I have more information. The rest of problems which have been reported are most probably due to the prematurely exited hanged process. Either starting in a new folder or passing --clean parameter can solve the issue.

Biobtree can work in local machine. This can be especially useful when large number of requests needs to be performed. For instance currently similar Uniprot tool documentation suggests either split the requests or download underlying data when number of requests are above 50K. These types of limitations for bulk requests are sensible for fair usage of a public service and can be more suitable with Biobtree type locally runnable tool.

Users custom dataset can be integrated.

Tool provides new intuitive web interface.