1 Introduction

Research organizations generate, manage, and use more and more knowledge resources which can be highly heterogenous in their origin, their scope, and their structure. Making this knowledge compliant to F.A.I.R. (Findable, Accessible, Interoperable, Reusable) principles is critical for facilitating the generation of new insights leveraging it. The aim of the TKCat (Tailored Knowledge Catalog) R package is to facilitate the management of such resources that are frequently used alone or in combination in research environments.

In TKCat, knowledge resources are manipulated as modeled database (MDB) objects. These objects provide access to the data tables along with a general description of the resource and a detail data model generated with ReDaMoR documenting the tables, their fields and their relationships. These MDB are then gathered in catalogs that can be easily explored an shared. TKCat provides tools to easily subset, filter and combine MDBs and create new catalogs suited for specific needs.

Currently, there are 3 different implementations of MDBs which are supported by TKCat: in R memory (memoMDB), in files (fileMDB) and in ClickHouse (chMDB).

This is document is divided in four main sections:

  • The first one describes how to build an MDB object, starting with a minimal example

  • The second section shows how to interact with MDB objects to extract and combine information of interest

  • The third section focuses on the use of the ClickHouse implementation of MDB (chMDB)

  • The fourth section corresponds to appendices providing technical information regarding ClickHouse related admin tasks and the implementation of collections which are used to identify and leverage potential relationships between different MDBs.

2 Create an MDB: a minimal example

This section shows how to create an MDB object starting from a set of tables in three steps:

  • Create a data model
  • Create and validate a modeled database (MDB) by binding the data model to the dataset
  • Document concept collections that can be used to make bridges across different MDBs

This example focuses on the Human Phenotype Ontology (HPO). The HPO aims to provide a standardized vocabulary of phenotypic abnormalities encountered in human diseases (Köhler et al. 2019).

2.1 Loading example data

A subset of the HPO is provided within the ReDaMoR package. We can read some of the tables as follow:

library(readr)
hpo_data_dir <- system.file("examples/HPO-subset", package="ReDaMoR")

The HPO_hp table gathers human phenotype identifiers, names and descriptions:

HPO_hp <- readr::read_tsv(
   file.path(hpo_data_dir, "HPO_hp.txt")
)
HPO_hp
## # A tibble: 500 × 4
##    id      name                                              description   level
##    <chr>   <chr>                                             <chr>         <dbl>
##  1 0000002 Abnormality of body height                        Deviation fr…     3
##  2 0000009 Functional abnormality of the bladder             Dysfunction …     6
##  3 0000014 Abnormality of the bladder                        An abnormali…     5
##  4 0000017 Nocturia                                          Abnormally i…     7
##  5 0000019 Urinary hesitancy                                 Difficulty i…     7
##  6 0000021 Megacystis                                        Dilatation o…     8
##  7 0000022 Abnormality of male internal genitalia            An abnormali…     6
##  8 0000024 Prostatitis                                       The presence…     8
##  9 0000025 Functional abnormality of male internal genitalia <NA>              6
## 10 0000030 Testicular gonadoblastoma                         The presence…     9
## # … with 490 more rows

The HPO_diseases table gathers disease identifiers and labels from different disease database.

HPO_diseases <- readr::read_tsv(
   file.path(hpo_data_dir, "HPO_diseases.txt")
)
HPO_diseases
## # A tibble: 1,903 × 3
##    db           id label                                                        
##    <chr>     <dbl> <chr>                                                        
##  1 DECIPHER     15 NF1-microdeletion syndrome                                   
##  2 DECIPHER     45 Xq28 (MECP2) duplication                                     
##  3 DECIPHER     65 ATR-16 syndrome                                              
##  4 OMIM     100050 AARSKOG SYNDROME, AUTOSOMAL DOMINANT                         
##  5 OMIM     100650 ALDEHYDE DEHYDROGENASE 2 FAMILY                              
##  6 OMIM     101800 ACRODYSOSTOSIS 1, WITH OR WITHOUT HORMONE RESISTANCE; ACRDYS1
##  7 OMIM     102500 HAJDU-CHENEY SYNDROME; HJCYS                                 
##  8 OMIM     102510 ACROPECTOROVERTEBRAL DYSPLASIA, F-FORM OF                    
##  9 OMIM     102700 SEVERE COMBINED IMMUNODEFICIENCY, AUTOSOMAL RECESSIVE, T CEL…
## 10 OMIM     102800 ADENOSINE TRIPHOSPHATASE DEFICIENCY, ANEMIA DUE TO           
## # … with 1,893 more rows

The HPO_diseaseHP table indicates which phenotype is triggered by each disease.

HPO_diseaseHP <- readr::read_tsv(
   file.path(hpo_data_dir, "HPO_diseaseHP.txt")
)
HPO_diseaseHP
## # A tibble: 2,594 × 3
##    db           id hp     
##    <chr>     <dbl> <chr>  
##  1 ORPHA    140976 0000002
##  2 ORPHA       432 0000002
##  3 DECIPHER     45 0000009
##  4 OMIM     300076 0000009
##  5 ORPHA    100996 0000009
##  6 ORPHA    100997 0000009
##  7 ORPHA      2571 0000009
##  8 ORPHA    391487 0000009
##  9 ORPHA    488594 0000009
## 10 ORPHA     71211 0000009
## # … with 2,584 more rows

2.2 Creating a data model with ReDaMoR

The ReDaMoR package can be used for drafting a data model from a set of table:

mhpo_dm <- ReDaMoR::df_to_model(HPO_hp, HPO_diseases, HPO_diseaseHP)
if(igraph_available){
   mhpo_dm %>%
      ReDaMoR::auto_layout(lengthMultiplier=80) %>% 
      plot()
}else{
   mhpo_dm %>%
      plot()
}

This data model is minimal: only the name of the tables, their fields and their types are documented. There is no additional constrain regarding the uniqueness or the completeness of the fields. Also there is no information regarding the relationships between the different tables. The model_relational_data() can be used to improve the documentation of the dataset according to what we know about it. This function raises a graphical interface for manipulating and modifying the data model (see ReDaMoR documentation).

mhpo_dm <- ReDaMoR::model_relational_data(mhpo_dm)

Below is the model we get after completing it using the function above.

plot(mhpo_dm)

In this model, we can see that:

  • id is the primary key of the HPO_hp table, and therefore this field must be unique;
  • db/id form the primary key of the HPO_diseases table and must also be unique when taken together;
  • all the fields excepted description (in the HPO_hp table) are complete (they cannot be NA);
  • the HPO_diseaseHP table refers to the HPO_hp table using its HPO_hp fields and to the HPO_diseases table using its db and id fields (such details are shown when putting cursor over the edges).

Moreover, some comments are added at the table and at the field level to give a better understanding of the data (shown when putting the cursor over the tables).

2.3 Binding the model to the data in an MDB object

The data model can be explicitly bound to the data in an MDB (Modeled DataBase) object as shown below. However, when trying to build the object with the tables we’ve read and the data model we have edited, we get the following error message.

mhpo_db <- memoMDB(
   dataTables=list(
      HPO_hp=HPO_hp, HPO_diseases=HPO_diseases, HPO_diseaseHP=HPO_diseaseHP
   ),
   dataModel=mhpo_dm,
   dbInfo=list(name="miniHPO")
)

miniHPO

FAILURE

Check configuration

  • Optional checks: unique, not nullable, foreign keys
  • Maximum number of records: Inf

HPO_hp

FAILURE

Field issues or warnings
  • description: SUCCESS Missing values 117/500 = 23%
  • level: FAILURE Unexpected “numeric”

HPO_diseases

FAILURE

Field issues or warnings
  • id: FAILURE Unexpected “numeric”

HPO_diseaseHP

FAILURE

Field issues or warnings
  • id: FAILURE Unexpected “numeric”

Indeed, according to the edited model (not the very first one automatically created by ReDaMoR), the HPO_hp$level field should contain integer values and the HPO_diseases$id and HPO_diseaseHP$id fields should contain character values. The type of the data is among the data model features that are automatically checked when building an MDB object (along with uniqueness or NA values for example).

To avoid this error, we can either change the type of the columns of the data tables:

HPO_hp <- mutate(HPO_hp, level=as.integer(level))
HPO_diseases <- mutate(HPO_diseases, id=as.character(id))
HPO_diseaseHP <- mutate(HPO_diseaseHP, id=as.character(id))
mhpo_db <- memoMDB(
   dataTables=list(
      HPO_hp=HPO_hp, HPO_diseases=HPO_diseases, HPO_diseaseHP=HPO_diseaseHP
   ),
   dataModel=mhpo_dm,
   dbInfo=list(name="miniHPO")
)

Or we can use the data model to read the data in a fileMDB object:

f_mhpo_db <- read_fileMDB(
   path=hpo_data_dir,
   dbInfo=list(name="miniHPO"),
   dataModel=mhpo_dm
)
## miniHPO
## SUCCESS
## 
## Check configuration
##    - Optional checks: 
##    - Maximum number of records: 10

The read_fileMDB() function identifies the text files to read in path according to the dataModel. It uses the types documented in the data model to read the files. By default, the field delimiter is \t, but another can be defined by writing a delim slot in the dbInfo parameter (e.g. dbInfo=list(name="miniHPO", delim="\t")).

As shown in the message above, by default, read_fileMDB() does not perform optional checks (unique fields, not nullable fields, foreign keys) and it only checks data on the 10 first records. Also, the fileMDB data are not loaded in memory until requested by the user. The object is then smaller than the memoMDB object even if they gather the same information.

print(object.size(mhpo_db), units="Kb")
## 691.9 Kb
print(object.size(f_mhpo_db), units="Kb")
## 23.5 Kb
compare_MDB(former=mhpo_db, new=f_mhpo_db) %>% 
   DT::datatable(
      rownames=FALSE,
      width="75%",
      options=list(dom="t", pageLength=nrow(.))
   )

2.4 Adding information about an MDB

In the table above we can see that several pieces of information are expected in an MDB object even if not mandatory (title, description, url, version, maintainer, timestamp). They can be provided in the dbInfo parameter of the MDB creator function (e.g. memoMDB()) or added afterward:

  • title, description and url are used to give more details about the scope of the data and their origin.
db_info(mhpo_db)$title <- "Very small extract of the human phenotype ontology"
db_info(mhpo_db)$description <- "For demonstrating ReDaMoR and TKCat capabilities, a very few information from the HPO (human phenotype ontology) has been extracted"
db_info(mhpo_db)$url <- "https://hpo.jax.org/app/"
  • version and maintainer are related to db information and the data model whereas timestamp should be used to document the data themselves.
db_info(mhpo_db)$version <- "0.1"
db_info(mhpo_db)$maintainer <- "Patrice Godard"
db_info(mhpo_db)$timestamp <- Sys.time()

All this information is displayed when printing the object:

mhpo_db
## memoMDB miniHPO (version 0.1, Patrice Godard): Very small extract of the human phenotype ontology
##    - 3 tables with 10 fields
## 
## No collection member
## 
## For demonstrating ReDaMoR and TKCat capabilities, a very few information from the HPO (human phenotype ontology) has been extracted
## (https://hpo.jax.org/app/)
## 
## Timesamp: 2023-02-16 13:25:50
## 

2.5 Documenting collection members

In the HPO example, one table regards human phenotypes (HPO_hp) and another human diseases (HPO_diseases). These concepts are general and referenced in many other knowledge or data resources (e.g. database providing information about disease genetics). Therefore, documenting formally such concepts will help to identify how to connect the HPO example to other resources referencing the same or related concepts.

In TKCat, these central concepts are referred as members of collections. Collections are pre-defined and members must be documented according to this definition. There are currently two collections provided within the TKCat package:

list_local_collections()
## # A tibble: 2 × 2
##   title     description                                  
##   <chr>     <chr>                                        
## 1 BE        Collection of biological entity (BE) concepts
## 2 Condition Collection of condition concepts

Additional collections can be defined by users according to their needs. Further information about collections implementation is provided in the appendix.

So far, there is no collection member documented in the HPO example described above, as indicated by the “No collection member” statement displayed when printing the object:

mhpo_db
## memoMDB miniHPO (version 0.1, Patrice Godard): Very small extract of the human phenotype ontology
##    - 3 tables with 10 fields
## 
## No collection member
## 
## For demonstrating ReDaMoR and TKCat capabilities, a very few information from the HPO (human phenotype ontology) has been extracted
## (https://hpo.jax.org/app/)
## 
## Timesamp: 2023-02-16 13:25:50
## 

However, as just discussed, the HPO_hp table refers to human phenotypes and the HPO_diseases table to human diseases. These concept corresponds to conditions and those tables can be documented as member of the Condition collection.

Condition members are documented calling the add_collection_member() function on the MDB object. The two other main arguments are the name of the collection and the name of the table in the MDB object. The other arguments to be provided depend on the collection. For Condition members, three additional arguments must be provided:

  • condition indicate the type of the condition (“Phenotype” or “Disease”)
  • source a reference source of the condition identifier
  • identifier a condition identifier

The functions get_local_collection() and show_collection_def() can be used together to identify valid arguments:

get_local_collection("Condition") %>%
   show_collection_def()
## Condition collection: Collection of condition concepts
## Arguments (non-mandatory arguments are between parentheses):
##    - condition:
##       + static: logical
##       + value: character
##    - source:
##       + static: logical
##       + value: character
##    - identifier:
##       + static: logical
##       + value: character

When calling add_collection_member(), these arguments must be provided as a list with 2 elements named “value” (a character) and “static” (a logical). If “static” is TRUE, “value” corresponds to the information shared by all the rows of the table. If “static” is FALSE, “value” indicates the name of the column which provides this information for each row.

The example below shows how the HPO_hp table is documented as a member of the Condition collection.

mhpo_db$HPO_hp
## # A tibble: 500 × 4
##    id      name                                              description   level
##    <chr>   <chr>                                             <chr>         <int>
##  1 0000002 Abnormality of body height                        Deviation fr…     3
##  2 0000009 Functional abnormality of the bladder             Dysfunction …     6
##  3 0000014 Abnormality of the bladder                        An abnormali…     5
##  4 0000017 Nocturia                                          Abnormally i…     7
##  5 0000019 Urinary hesitancy                                 Difficulty i…     7
##  6 0000021 Megacystis                                        Dilatation o…     8
##  7 0000022 Abnormality of male internal genitalia            An abnormali…     6
##  8 0000024 Prostatitis                                       The presence…     8
##  9 0000025 Functional abnormality of male internal genitalia <NA>              6
## 10 0000030 Testicular gonadoblastoma                         The presence…     9
## # … with 490 more rows
mhpo_db <- add_collection_member(
   mhpo_db, collection="Condition", table="HPO_hp",
   condition=list(value="Phenotype", static=TRUE),
   source=list(value="HP", static=TRUE),
   identifier=list(value="id", static=FALSE)
)

All rows in this table correspond to a condition of type “Phenotype” (condition=list(value="Phenotype", static=TRUE)). The phenotype identifiers are all taken from the same source, “HP” (source=list(value="HP", static=TRUE)). The phenotype identifiers are provided in the “id” column of the table (identifier=list(value="id", static=FALSE)).

The example below shows how the HPO_disease table is documented also as a member of the Condition collection. In this case, the source of disease identifier can be different from one row to the other and is provided in the “db” column (source=list(value="db", static=FALSE)).

mhpo_db <- add_collection_member(
   mhpo_db, collection="Condition", table="HPO_diseases",
   condition=list(value="Disease", static=TRUE),
   source=list(value="db", static=FALSE),
   identifier=list(value="id", static=FALSE)
)

Now, the existence of collection members is shown when printing the MDB object:

mhpo_db
## memoMDB miniHPO (version 0.1, Patrice Godard): Very small extract of the human phenotype ontology
##    - 3 tables with 10 fields
## 
## Collection members: 
##    - 2 Condition members
## 
## For demonstrating ReDaMoR and TKCat capabilities, a very few information from the HPO (human phenotype ontology) has been extracted
## (https://hpo.jax.org/app/)
## 
## Timesamp: 2023-02-16 13:25:50
## 

And the documented collection members of an MDB can be displayed as following:

collection_members(mhpo_db)
## # A tibble: 6 × 9
##   collection cid                   resource   mid table field static value type 
##   <chr>      <chr>                 <chr>    <int> <chr> <chr> <lgl>  <chr> <chr>
## 1 Condition  miniHPO_Condition_1.0 miniHPO      1 HPO_… cond… TRUE   Phen… <NA> 
## 2 Condition  miniHPO_Condition_1.0 miniHPO      1 HPO_… sour… TRUE   HP    <NA> 
## 3 Condition  miniHPO_Condition_1.0 miniHPO      1 HPO_… iden… FALSE  id    <NA> 
## 4 Condition  miniHPO_Condition_1.0 miniHPO      2 HPO_… cond… TRUE   Dise… <NA> 
## 5 Condition  miniHPO_Condition_1.0 miniHPO      2 HPO_… sour… FALSE  db    <NA> 
## 6 Condition  miniHPO_Condition_1.0 miniHPO      2 HPO_… iden… FALSE  id    <NA>

The use of collection members to link or integrate different MDBs will be described later in this document

2.6 Writing an MDB in files

Once an MDB has been created and documented in can be written in a directory:

tmpDir <- tempdir()
as_fileMDB(mhpo_db, path=tmpDir, htmlModel=FALSE)

The structure of the created directory is the following:

## miniHPO                                         
##  ¦--DESCRIPTION.json                            
##  ¦--data                                        
##  ¦   ¦--HPO_diseaseHP.txt.gz                    
##  ¦   ¦--HPO_diseases.txt.gz                     
##  ¦   °--HPO_hp.txt.gz                           
##  °--model                                       
##      ¦--Collections                             
##      ¦   °--Condition-miniHPO_Condition_1.0.json
##      °--miniHPO.json

All the data are in the data folder whereas the data model and collection members are written in json files in the model folder. The DESCRIPTION.json file gather db information and information about how to read the data files (i.e. delim, na).

This folder can be shared and it’s then easy to get all the data and the corresponding documentation from it back in R:

read_fileMDB(file.path(tmpDir, "miniHPO"))
## miniHPO
## SUCCESS
## 
## Check configuration
##    - Optional checks: 
##    - Maximum number of records: 10
## fileMDB miniHPO (version 0.1, Patrice Godard): Very small extract of the human phenotype ontology
##    - 3 tables with 10 fields
## 
## Collection members: 
##    - 2 Condition members
## 
## For demonstrating ReDaMoR and TKCat capabilities, a very few information from the HPO (human phenotype ontology) has been extracted
## (https://hpo.jax.org/app/)
## 
## Timesamp: 2023-02-16 13:25:50
## 

Also writing these data and related information in text files make them convenient to share with people using them in other analytical environments than R.

3 Leveraging MDB

The former section showed how to create and save an MDB object. This section describes how MDBs can be used, filtered and combined to efficiently leverage their content.

As a reminder, a modeled database (MDB) in TKCat gathers the following information:

  • General database information including a mandatory name and optionally the following fields: title, description, url, version and maintainer.
  • A ReDaMoR data model.
  • A list of tables corresponding to reference concepts shared by different MDBs. The way these concepts are identified is defined in specific documents called collections.
  • The data themselves organized according to the data model.

3.1 Loading example data

To illustrate how MDBs can be used, some example data are provided within the ReDaMoR and the TKCat package. The following paragraphs show how to load them in the R session.

3.1.1 HPO

A subset of the Human Phenotype Ontology (HPO) is provided within the ReDaMoR package. The HPO aims to provide a standardized vocabulary of phenotypic abnormalities encountered in human diseases (Köhler et al. 2019). An MDB object based on files (see MDB implementations) can be read as shown below. As explained above, the data provided by the path parameter are documented with a model (dataModel parameter) and general information (dbInfo parameter).

file_hpo <- read_fileMDB(
   path=system.file("examples/HPO-subset", package="ReDaMoR"),
   dataModel=system.file("examples/HPO-model.json", package="ReDaMoR"),
   dbInfo=list(
      "name"="HPO",
      "title"="Data extracted from the HPO database",
      "description"=paste(
         "This is a very small subset of the HPO!",
         "Visit the reference URL for more information."
      ),
      "url"="http://human-phenotype-ontology.github.io/"
   )
)
## HPO
## SUCCESS
## 
## Check configuration
##    - Optional checks: 
##    - Maximum number of records: 10

The message displayed in the console indicates if the data fit the data model. It relies on the ReDaMoR::confront_data() functions and check by default the first 10 rows of each file.

The data model can then be drawn.

plot(data_model(file_hpo))

The data model shows that this MDB contains the 3 tables taken into account in the minimal example. The additional tables provides mainly supplementary details regarding phenotype and diseases. Still, the HPO_hp and the HPO_disease table are members of the Condition collection and can be documented as such, as explained above.

file_hpo <- file_hpo %>% 
   add_collection_member(
      collection="Condition", table="HPO_hp",
      condition=list(value="Phenotype", static=TRUE),
      source=list(value="HP", static=TRUE),
      identifier=list(value="id", static=FALSE)
   ) %>% 
   add_collection_member(
      collection="Condition", table="HPO_diseases",
      condition=list(value="Disease", static=TRUE),
      source=list(value="db", static=FALSE),
      identifier=list(value="id", static=FALSE)
   )

3.1.2 ClinVar

A subset of the ClinVar database is provided within this package. ClinVar is a freely accessible, public archive of reports of the relationships among human variations and phenotypes, with supporting evidence (Landrum et al. 2018). This resource can be read as a fileMDB as shown above. However, in this case all the documenting information is included in the resource directory, making it easier to read as explained above.

file_clinvar <- read_fileMDB(
   path=system.file("examples/ClinVar", package="TKCat")
)
## ClinVar
## SUCCESS
## 
## Check configuration
##    - Optional checks: 
##    - Maximum number of records: 10
file_clinvar
## fileMDB ClinVar (version 0.9, Patrice Godard <patrice.godard@ucb.com>): Data extracted from the ClinVar database
##    - 21 tables with 86 fields
## 
## Collection members: 
##    - 1 BE member
##    - 2 Condition members
## 
## ClinVar is a freely accessible, public archive of reports of the relationships among human variations and phenotypes, with supporting evidence. This is a very small subset of ClinVar! Visit the reference URL for more information.
## (https://www.ncbi.nlm.nih.gov/clinvar/)
## 
## 

3.1.3 CHEMBL

Similarly, a self-documented subset of the CHEMBL database is also provided in the TKCat package. It can be read the same way.

file_chembl <- read_fileMDB(
   path=system.file("examples/CHEMBL", package="TKCat")
)
## CHEMBL
## SUCCESS
## 
## Check configuration
##    - Optional checks: 
##    - Maximum number of records: 10

CHEMBL is a manually curated chemical database of bioactive molecules with drug-like properties (Mendez et al. 2019).

file_chembl
## fileMDB CHEMBL (version 0.2, Liesbeth François <liesbeth.francois@ucb.com>): Data extracted from the CHEMBL database
##    - 10 tables with 61 fields
## 
## Collection members: 
##    - 1 BE member
##    - 1 Condition member
## 
## CHEMBL is a manually curated chemical database of bioactive molecules with drug-like properties. This is a very small subset of CHEMBL! Visit the reference URL for more information.
## (https://www.ebi.ac.uk/chembl/)
## 
## 

3.2 MDB implementations

There are 3 main implementations of MDBs:

  • fileMDB objects keep the data in files and load them only when requested by the user. These implementation is the first one which is used when reading MDB as demonstrated in the examples above.

  • memoMDB objects have all the data loaded in memory. These objects are very easy to use but can take time to load and can use a lot of memory.

  • chMDB objects get the data from a ClickHouse database providing a catalog of MDBs as described in the dedicated section.

The different implementations can be converted to each others using as_fileMDB(), as_memoMDB() and as_chMDB() functions.

memo_clinvar <- as_memoMDB(file_clinvar)
object.size(file_clinvar) %>% print(units="Kb")
## 155.2 Kb
object.size(memo_clinvar) %>% print(units="Kb")
## 760.5 Kb

A fourth implementation is metaMDB which combines several MDBs glued together with relational tables (see the Merging with collections part).

Most of the functions described below work with any MDB implementation, and a few functions are specific to each implementation.

3.3 Exploring information

General information can be retrieved (and potentialy updated) using the db_info() function.

db_info(file_clinvar)
## $name
## [1] "ClinVar"
## 
## $title
## [1] "Data extracted from the ClinVar database"
## 
## $description
## [1] "ClinVar is a freely accessible, public archive of reports of the relationships among human variations and phenotypes, with supporting evidence. This is a very small subset of ClinVar! Visit the reference URL for more information."
## 
## $url
## [1] "https://www.ncbi.nlm.nih.gov/clinvar/"
## 
## $version
## [1] "0.9"
## 
## $maintainer
## [1] "Patrice Godard <patrice.godard@ucb.com>"
## 
## $timestamp
## [1] NA

As shown above the data model of an MDB can be retrieved and plot the following way.

plot(data_model(file_clinvar))

Tables names can be listed with the names() function and potentially renamed with names()<- or rename() functions (the tables have been renamed here to improve the readability of the following examples).

names(file_clinvar)
##  [1] "ClinVar_ReferenceClinVarAssertion" "ClinVar_rcvaVariant"              
##  [3] "ClinVar_ClinVarAssertions"         "ClinVar_rcvaInhMode"              
##  [5] "ClinVar_rcvaObservedIn"            "ClinVar_rcvaTraits"               
##  [7] "ClinVar_clinSigOrder"              "ClinVar_revStatOrder"             
##  [9] "ClinVar_variants"                  "ClinVar_cvaObservedIn"            
## [11] "ClinVar_cvaSubmitters"             "ClinVar_traits"                   
## [13] "ClinVar_varEntrez"                 "ClinVar_varAttributes"            
## [15] "ClinVar_varCytoLoc"                "ClinVar_varNames"                 
## [17] "ClinVar_varSeqLoc"                 "ClinVar_varXRef"                  
## [19] "ClinVar_traitCref"                 "ClinVar_traitNames"               
## [21] "ClinVar_entrezNames"
file_clinvar <- file_clinvar %>% 
   set_names(sub("ClinVar_", "", names(.))) 
names(file_clinvar)
##  [1] "ReferenceClinVarAssertion" "rcvaVariant"              
##  [3] "ClinVarAssertions"         "rcvaInhMode"              
##  [5] "rcvaObservedIn"            "rcvaTraits"               
##  [7] "clinSigOrder"              "revStatOrder"             
##  [9] "variants"                  "cvaObservedIn"            
## [11] "cvaSubmitters"             "traits"                   
## [13] "varEntrez"                 "varAttributes"            
## [15] "varCytoLoc"                "varNames"                 
## [17] "varSeqLoc"                 "varXRef"                  
## [19] "traitCref"                 "traitNames"               
## [21] "entrezNames"

The different collection members of an MDBs are listed with the collection_members() function.

collection_members(file_clinvar)
## # A tibble: 10 × 9
##    collection cid                   resou…¹   mid table field static value type 
##    <chr>      <chr>                 <chr>   <int> <chr> <chr> <lgl>  <chr> <chr>
##  1 Condition  ClinVar_conditions_1… ClinVar     2 trai… cond… TRUE   Dise… <NA> 
##  2 Condition  ClinVar_conditions_1… ClinVar     2 trai… iden… FALSE  id    <NA> 
##  3 Condition  ClinVar_conditions_1… ClinVar     2 trai… sour… TRUE   Clin… <NA> 
##  4 Condition  ClinVar_conditions_1… ClinVar     1 trai… cond… TRUE   Dise… <NA> 
##  5 Condition  ClinVar_conditions_1… ClinVar     1 trai… iden… FALSE  id    <NA> 
##  6 Condition  ClinVar_conditions_1… ClinVar     1 trai… sour… FALSE  db    <NA> 
##  7 BE         ClinVar_BE_1.0        ClinVar     1 entr… be    TRUE   Gene  <NA> 
##  8 BE         ClinVar_BE_1.0        ClinVar     1 entr… iden… FALSE  entr… <NA> 
##  9 BE         ClinVar_BE_1.0        ClinVar     1 entr… orga… TRUE   Homo… Scie…
## 10 BE         ClinVar_BE_1.0        ClinVar     1 entr… sour… TRUE   Entr… <NA> 
## # … with abbreviated variable name ¹​resource

The following functions are use to get the number of tables, the number of fields per table and the number of records.

length(file_clinvar)        # Number of tables
## [1] 21
lengths(file_clinvar)       # Number of fields per table
## ReferenceClinVarAssertion               rcvaVariant         ClinVarAssertions 
##                         8                         2                         4 
##               rcvaInhMode            rcvaObservedIn                rcvaTraits 
##                         2                         6                         3 
##              clinSigOrder              revStatOrder                  variants 
##                         2                         2                         3 
##             cvaObservedIn             cvaSubmitters                    traits 
##                         4                         3                         2 
##                 varEntrez             varAttributes                varCytoLoc 
##                         3                         5                         2 
##                  varNames                 varSeqLoc                   varXRef 
##                         3                        18                         4 
##                 traitCref                traitNames               entrezNames 
##                         4                         3                         3
count_records(file_clinvar) # Number of records per table
## ReferenceClinVarAssertion               rcvaVariant         ClinVarAssertions 
##                       166                       166                       409 
##               rcvaInhMode            rcvaObservedIn                rcvaTraits 
##                        16                       337                       166 
##              clinSigOrder              revStatOrder                  variants 
##                        11                         2                       138 
##             cvaObservedIn             cvaSubmitters                    traits 
##                       412                       416                        18 
##                 varEntrez             varAttributes                varCytoLoc 
##                       145                      2262                       138 
##                  varNames                 varSeqLoc                   varXRef 
##                       188                       280                       244 
##                 traitCref                traitNames               entrezNames 
##                        50                        44                        20

The count_records() function can take a lot of time when dealing with fileMDB objects if the data files are very large. In such case it could be more efficient to list data file size instead.

data_file_size(file_clinvar, hr=TRUE)
## # A tibble: 21 × 3
##    table                     size   compressed
##    <chr>                     <chr>  <lgl>     
##  1 ReferenceClinVarAssertion 4.6 KB TRUE      
##  2 rcvaVariant               947 B  TRUE      
##  3 ClinVarAssertions         4.2 KB TRUE      
##  4 rcvaInhMode               152 B  TRUE      
##  5 rcvaObservedIn            1.4 KB TRUE      
##  6 rcvaTraits                788 B  TRUE      
##  7 clinSigOrder              145 B  TRUE      
##  8 revStatOrder              101 B  TRUE      
##  9 variants                  2.1 KB TRUE      
## 10 cvaObservedIn             1.8 KB TRUE      
## # … with 11 more rows

3.4 Pulling, subsetting and combining

There are several possible ways to pull data tables from MDBs. The following lines return the same result displayed below (only once).

data_tables(file_clinvar, "traitNames")[[1]]
file_clinvar[["traitNames"]]
file_clinvar$"traitNames"
file_clinvar %>% pull(traitNames)
## # A tibble: 44 × 3
##     t.id name                                                              type 
##    <int> <chr>                                                             <chr>
##  1   912 Chudley-McCullough syndrome                                       Pref…
##  2   912 Deafness, autosomal recessive 82                                  Alte…
##  3   912 Deafness, bilateral sensorineural, and hydrocephalus due to fora… Alte…
##  4   912 Deafness, sensorineural, with partial agenesis of the corpus cal… Alte…
##  5  1352 CTSD-Related Neuronal Ceroid-Lipofuscinosis                       Alte…
##  6  1352 Ceroid lipofuscinosis neuronal Cathepsin D-deficient              Alte…
##  7  1352 Neuronal ceroid lipofuscinosis 10                                 Pref…
##  8  1352 Neuronal ceroid lipofuscinosis due to Cathepsin D deficiency      Alte…
##  9  1481 Diabetes mellitus, neonatal, with congenital hypothyroidism       Pref…
## 10  1481 NDH SYNDROME                                                      Alte…
## # … with 34 more rows

MDBs can also be subset and combined. The corresponding functions ensure that the data model is fulfilled by the data tables.

file_clinvar[1:3]
## fileMDB ClinVar (version 0.9, Patrice Godard <patrice.godard@ucb.com>): Data extracted from the ClinVar database
##    - 3 tables with 14 fields
## 
## No collection member
## 
## ClinVar is a freely accessible, public archive of reports of the relationships among human variations and phenotypes, with supporting evidence. This is a very small subset of ClinVar! Visit the reference URL for more information.
## (https://www.ncbi.nlm.nih.gov/clinvar/)
## 
## 
if(igraph_available){
   c(file_clinvar[1:3], file_hpo[c(1,5,7)]) %>% 
      data_model() %>% auto_layout(force=TRUE) %>% plot()
}else{
   c(file_clinvar[1:3], file_hpo[c(1,5,7)]) %>% 
      data_model() %>% plot()
}

The function c() concatenates the provided MDB after checking that tables names are not duplicated. It does not integrate the data with any relational table. This can achieved by merging the MDBs as described in the Merging with collections section.

3.5 Filtering and joining

An MDB can be filtered by filtering one or several tables based on field values. The filtering is propagated to other tables using the embedded data model.

In the example below, the file_clinvar object is filtered in order to focus on a few genes with pathogenic variants. The table below compares the number of rows before (“ori”) and after (“filt”) filtering.

filtered_clinvar <- file_clinvar %>%
   filter(
      entrezNames = symbol %in% c("PIK3R2", "UGT1A8")
   ) %>% 
   slice(ReferenceClinVarAssertion=grep(
      "pathogen",
      .$ReferenceClinVarAssertion$clinicalSignificance,
      ignore.case=TRUE
   ))
left_join(
   dims(file_clinvar) %>% select(name, nrow),
   dims(filtered_clinvar) %>% select(name, nrow),
   by="name",
   suffix=c("_ori", "_filt")
)
## # A tibble: 21 × 3
##    name                      nrow_ori nrow_filt
##    <chr>                        <dbl>     <int>
##  1 ReferenceClinVarAssertion      166         4
##  2 rcvaVariant                    166         4
##  3 ClinVarAssertions              409        15
##  4 rcvaInhMode                     16         0
##  5 rcvaObservedIn                 337        10
##  6 rcvaTraits                     166         4
##  7 clinSigOrder                    11         3
##  8 revStatOrder                     2         1
##  9 variants                       138         3
## 10 cvaObservedIn                  412        15
## # … with 11 more rows

The object returned by filter() or slice is a memoMDB: all the data are in memory.

Tables can be easily joined to get diseases associated to the genes of interest in a single table as shown below.

gene_traits <- filtered_clinvar %>% 
   join_mdb_tables(
      "entrezNames", "varEntrez", "variants", "rcvaVariant",
      "ReferenceClinVarAssertion", "rcvaTraits", "traits"
   )
gene_traits$entrezNames %>%
   select(symbol, name, variants.type, variants.name, traitType, traits.name)
## # A tibble: 4 × 6
##   symbol name                                    varia…¹ varia…² trait…³ trait…⁴
##   <chr>  <chr>                                   <chr>   <chr>   <chr>   <chr>  
## 1 PIK3R2 phosphoinositide-3-kinase regulatory s… single… NM_005… Disease Megale…
## 2 PIK3R2 phosphoinositide-3-kinase regulatory s… single… NM_005… Disease not pr…
## 3 PIK3R2 phosphoinositide-3-kinase regulatory s… single… NM_005… Disease not pr…
## 4 UGT1A8 UDP glucuronosyltransferase family 1 m… Micros… UGT1A1… Disease Gilber…
## # … with abbreviated variable names ¹​variants.type, ²​variants.name, ³​traitType,
## #   ⁴​traits.name

3.6 Merging MDBs with collections

Until now, we have seen how to use individual MDB by exploring general information about it, extracting tables, filtering and joining data. This part shows how to use collections to identify relationships between MDBs and to leverage these relationships to integrate them. Documenting collection members has been described above and further information about collections implementation is provided in the appendix.

3.6.1 Collections and collection members

As explained above, some databases refer to the same concepts and could be integrated accordingly. However they often use different vocabularies.

For example, both CHEMBL and ClinVar refer to biological entities (BE) for documenting drug targets or disease causal genes. CHEMBL refers to drug target in the CHEMBL_component_sequence table using mainly Uniprot peptide identifiers from different species.

file_chembl$CHEMBL_component_sequence
## # A tibble: 35 × 5
##    component_id accession organism                  db_source db_version
##           <int> <chr>     <chr>                     <chr>     <chr>     
##  1          259 P15260    Homo sapiens              Uniprot   2019_09   
##  2          327 Q99062    Homo sapiens              Uniprot   2019_09   
##  3          752 P35563    Rattus norvegicus         Uniprot   2019_09   
##  4          917 P07339    Homo sapiens              Uniprot   2019_09   
##  5         1807 Q54A96    Plasmodium falciparum     Uniprot   2019_09   
##  6         2180 P67774    Bos taurus                Uniprot   2019_09   
##  7         2398 P25098    Homo sapiens              Uniprot   2019_09   
##  8         2541 Q8II92    Plasmodium falciparum 3D7 Uniprot   2019_09   
##  9         3803 Q64346    Rattus norvegicus         Uniprot   2019_09   
## 10         4395 O60502    Homo sapiens              Uniprot   2019_09   
## # … with 25 more rows

Whereas ClinVar refers to causal genes in the entrezNames table using human Entrez gene identifiers.

file_clinvar$entrezNames
## # A tibble: 20 × 3
##       entrez name                                                         symbol
##        <int> <chr>                                                        <chr> 
##  1      1509 cathepsin D                                                  CTSD  
##  2      1903 sphingosine-1-phosphate receptor 3                           S1PR3 
##  3      3300 DnaJ heat shock protein family (Hsp40) member B2             DNAJB2
##  4      3423 iduronate 2-sulfatase                                        IDS   
##  5      3910 laminin subunit alpha 4                                      LAMA4 
##  6      5296 phosphoinositide-3-kinase regulatory subunit 2               PIK3R2
##  7      6748 signal sequence receptor subunit 4                           SSR4  
##  8      7633 zinc finger protein 79                                       ZNF79 
##  9     22906 trafficking kinesin protein 1                                TRAK1 
## 10     23155 chloride channel CLIC like 1                                 CLCC1 
## 11     26251 potassium voltage-gated channel modifier subfamily G member… KCNG2 
## 12     29851 inducible T cell costimulator                                ICOS  
## 13     54576 UDP glucuronosyltransferase family 1 member A8               UGT1A8
## 14     57684 zinc finger and BTB domain containing 26                     ZBTB26
## 15    115948 outer dynein arm docking complex subunit 3                   ODAD3 
## 16    139716 GRB2 associated binding protein 3                            GAB3  
## 17    169792 GLIS family zinc finger 3                                    GLIS3 
## 18    407054 microRNA 98                                                  MIR98 
## 19    441531 phosphoglycerate mutase family member 4                      PGAM4 
## 20 105373557 serous ovarian cancer associated RNA                         SOCAR

Since peptides are coded by genes, there is a biological relationship between these two types of BE, and several tools exist to convert such BE identifiers from one scope to the other (e.g. BED (Godard and Eyll 2018), mygene (Wu, MacLeod, and Su 2012), biomaRt (Kinsella et al. 2011)).

TKCat provides mechanism to document these scopes in order to allow automatic conversions from and to any of them. Those concepts are called Collections in TKCat and they should be formally defined before being able to document any of their members. Two collection definitions are provided within the TKCat package and other can be imported with the import_local_collection() function.

list_local_collections()
## # A tibble: 2 × 2
##   title     description                                  
##   <chr>     <chr>                                        
## 1 BE        Collection of biological entity (BE) concepts
## 2 Condition Collection of condition concepts

Here are the definition of the BE collection members provided by the CHEMBL_component_sequence and the entrezNames tables.

collection_members(file_chembl, "BE")
## # A tibble: 4 × 9
##   collection cid           resource   mid table         field static value type 
##   <chr>      <chr>         <chr>    <int> <chr>         <chr> <lgl>  <chr> <chr>
## 1 BE         CHEMBL_BE_1.0 CHEMBL       1 CHEMBL_compo… be    TRUE   Pept… <NA> 
## 2 BE         CHEMBL_BE_1.0 CHEMBL       1 CHEMBL_compo… iden… FALSE  acce… <NA> 
## 3 BE         CHEMBL_BE_1.0 CHEMBL       1 CHEMBL_compo… sour… FALSE  db_s… <NA> 
## 4 BE         CHEMBL_BE_1.0 CHEMBL       1 CHEMBL_compo… orga… FALSE  orga… Scie…
collection_members(file_clinvar, "BE")
## # A tibble: 4 × 9
##   collection cid            resource   mid table       field  static value type 
##   <chr>      <chr>          <chr>    <int> <chr>       <chr>  <lgl>  <chr> <chr>
## 1 BE         ClinVar_BE_1.0 ClinVar      1 entrezNames be     TRUE   Gene  <NA> 
## 2 BE         ClinVar_BE_1.0 ClinVar      1 entrezNames ident… FALSE  entr… <NA> 
## 3 BE         ClinVar_BE_1.0 ClinVar      1 entrezNames organ… TRUE   Homo… Scie…
## 4 BE         ClinVar_BE_1.0 ClinVar      1 entrezNames source TRUE   Entr… <NA>

The Collection column indicates the collection to which the table refers. The cid column indicates the version of the collection definition which should correspond to the $id of JSON schema. The resource column indicates the name of the resource and the mid column an identifier which is unique for each member of a collection in each resource. The field column indicates each part of the scope of collection. In the case of BE, 4 fields should be documented:

  • be: the type of BE (e.g. Gene or Peptide)
  • source: the source of the identifier (e.g. EntrezGene or Peptide)
  • organism: the organism to which the identifier refers (e.g Homo sapiens)
  • identifier: the identifier itself.

Each of these fields can be static or not. TRUE means that the value of this field is the same for all the records and is provided in the value column. Whereas FALSE means that the value can be different for each record and is provided in the column the name of which is given in the value column. The type column is only used for the organism field in the case of the BE collection and can take 2 values: “Scientific name” or “NCBI taxon identifier”. The definition of the pre-build BE collection members follows the terminology used in the BED package (Godard and Eyll 2018). But it can be adapted according to the solution chosen for converting BE identifiers from one scope to another.

Setting up the definition of such scope is done using the add_collection_member() function as shown above in the minimal example and in the Reading HPO example.

3.6.2 Shared collections and merging

The aim of collections is to identify potential bridges between MDBs. The get_shared_collection() function is used to list all the collections shared by two MDBs.

get_shared_collections(filtered_clinvar, file_chembl)
## # A tibble: 3 × 5
##   collection table.x     mid.x table.y                   mid.y
##   <chr>      <chr>       <int> <chr>                     <int>
## 1 Condition  traits          2 CHEMBL_drug_indication        1
## 2 Condition  traitCref       1 CHEMBL_drug_indication        1
## 3 BE         entrezNames     1 CHEMBL_component_sequence     1

In this example, there are 3 different ways to merge the two MDBs filtered_clinvar and file_chembl:

  • Based on conditions provided respectively in the traits and in the CHEMBL_drug_indication tables
  • Based on conditions provided respectively in the traitsCref and in the CHEMBL_drug_indication tables
  • Based on BE provided respectively in the entrezNames and in the CHEMBL_component_sequence tables

The code below shows how to merge these two resources based on BE information. To achieve this task it relies on a function provided with TKCat along with BE collection definition (to get the function: get_collection_mapper("BE")). This function uses the BED package (Godard and Eyll 2018) and you need this package to be installed with a connection to BED database in order to run the code below.

bedCheck <- try(BED::checkBedConn())
## http://bel028272:5454
## BED
## UCB-Human-Internal
## 2023.02.11
## Cache ON
if(!inherits(bedCheck, "try-error") && bedCheck){
   sel_coll <- get_shared_collections(file_clinvar, file_chembl) %>% 
      filter(collection=="BE")
   filtered_cv_chembl <- merge(
      x=file_clinvar,
      y=file_chembl,
      by=sel_coll,
      dmAutoLayout=igraph_available
   )
}

The returned object is a metaMDB gathering the original MDBs and a relational table between members of the same collection as defined by the by parameter.

Additional information about collection can be found below in the appendix.

3.6.3 Merging without collection

If the collection column of the by parameter is NA, then the relational table is built by merging identical columns in table.x and table.y (No conversion occurs). For example, file_hpo and file_clinvar MDBs could be merged according to conditions provided in the HPO_diseases and the traitCref tables respectively.

get_shared_collections(file_hpo, file_clinvar)
## # A tibble: 4 × 5
##   collection table.x      mid.x table.y   mid.y
##   <chr>      <chr>        <int> <chr>     <int>
## 1 Condition  HPO_hp           1 traits        2
## 2 Condition  HPO_hp           1 traitCref     1
## 3 Condition  HPO_diseases     2 traits        2
## 4 Condition  HPO_diseases     2 traitCref     1

These conditions could be converted using a function provided with TKCat (get_collection_mapper("Condition")) and which rely on the DODO package (François, Eyll, and Godard 2020). The two tables can also be simply concatenated without applying any conversion (loosing the advantage of such conversion obviously).

sel_coll <- get_shared_collections(file_hpo, file_clinvar) %>% 
   filter(table.x=="HPO_diseases", table.y=="traitCref") %>% 
   mutate(collection=NA)
sel_coll
## # A tibble: 1 × 5
##   collection table.x      mid.x table.y   mid.y
##   <lgl>      <chr>        <int> <chr>     <int>
## 1 NA         HPO_diseases     2 traitCref     1

The merge() function gather the two MDBs in one metaMDB and create a association table based on the by argument. This association table (“HPO_diseases_traitCref”) is displayed in yellow in the data model of the created metaMDB as shown below.

hpo_clinvar <- merge(
   file_hpo, file_clinvar, by=sel_coll, dmAutoLayout=igraph_available
)
plot(data_model(hpo_clinvar))
hpo_clinvar$HPO_diseases_traitCref
## # A tibble: 1,950 × 2
##    db       id    
##    <chr>    <chr> 
##  1 DECIPHER 15    
##  2 DECIPHER 45    
##  3 DECIPHER 65    
##  4 OMIM     100050
##  5 OMIM     100650
##  6 OMIM     101800
##  7 OMIM     102500
##  8 OMIM     102510
##  9 OMIM     102700
## 10 OMIM     102800
## # … with 1,940 more rows

4 A centralized catalog of MDB in ClickHouse (chTKCat)

4.1 Local TKCat

MDB can be gathered in a TKCat (Tailored Knowledge Catalog) object.

k <- TKCat(file_hpo, file_clinvar)

Gathering MDBs in such a catalog facilitate their exploration and their preparation for potential integration. Several functions are available to achieve this goal.

list_MDBs(k)                     # list all the MDBs in a TKCat object
## # A tibble: 2 × 7
##   name    title                descr…¹ url   version maint…² timestamp
##   <chr>   <chr>                <chr>   <chr> <chr>   <chr>   <dttm>   
## 1 HPO     Data extracted from… This i… http… <NA>    <NA>    NA       
## 2 ClinVar Data extracted from… ClinVa… http… 0.9     Patric… NA       
## # … with abbreviated variable names ¹​description, ²​maintainer
get_MDB(k, "HPO")                # get a specific MDBs from the catalog
## fileMDB HPO: Data extracted from the HPO database
##    - 9 tables with 25 fields
## 
## Collection members: 
##    - 2 Condition members
## 
## This is a very small subset of the HPO! Visit the reference URL for more information.
## (http://human-phenotype-ontology.github.io/)
## 
## 
search_MDB_tables(k, "disease")  # Search table about "disease"
## # A tibble: 3 × 3
##   resource name                comment                 
##   <chr>    <chr>               <chr>                   
## 1 HPO      HPO_diseases        Diseases                
## 2 HPO      HPO_diseaseHP       HP presented by diseases
## 3 HPO      HPO_diseaseSynonyms Disease synonyms
search_MDB_fields(k, "disease")  # Search a field about "disease"
## # A tibble: 8 × 7
##   resource table               name    type      nullable unique comment        
##   <chr>    <chr>               <chr>   <chr>     <lgl>    <lgl>  <chr>          
## 1 HPO      HPO_diseases        db      character FALSE    FALSE  Disease databa…
## 2 HPO      HPO_diseases        id      character FALSE    FALSE  Disease ID     
## 3 HPO      HPO_diseases        label   character FALSE    FALSE  Disease lable …
## 4 HPO      HPO_diseaseHP       db      character FALSE    FALSE  Disease databa…
## 5 HPO      HPO_diseaseHP       id      character FALSE    FALSE  Disease ID     
## 6 HPO      HPO_diseaseSynonyms db      character FALSE    FALSE  Disease databa…
## 7 HPO      HPO_diseaseSynonyms id      character FALSE    FALSE  Disease ID     
## 8 HPO      HPO_diseaseSynonyms synonym character FALSE    FALSE  Disease synonym
collection_members(k)            # Get collection members of the different MDBs
## # A tibble: 5 × 3
##   resource collection table       
##   <chr>    <chr>      <chr>       
## 1 HPO      Condition  HPO_hp      
## 2 HPO      Condition  HPO_diseases
## 3 ClinVar  Condition  traits      
## 4 ClinVar  Condition  traitCref   
## 5 ClinVar  BE         entrezNames
c(k, TKCat(file_chembl))         # Merge 2 TKCat objects
## TKCat gathering 3 MDB objects

The function explore_MDBs() launches a shiny interface to explore MDBs in a TKCat object. This exploration interface can be easily deployed using an app.R file with content similar to the one below.

library(TKCat)
explore_MDBs(k, download=TRUE)

In this interface the users can explore the resources available in the catalog. They can browse the data model of each of them with some sample data. They can also search for information provided in resources, tables or fields. Finally, if the parameter download is set to TRUE, the users will also be able to download the data: either each table individually or an archive of the whole MDB.

4.2 chTKCat

A chTKCat object is a catalog of MDB as a TKCat object described above but relying on a ClickHouse database. This part focuses on using and querying a chTKCat object. The installation and the initialization of a ClickHouse database ready for TKCat are described below in the appendix.

The connection to the ClickHouse TKCat database is achieved using the chTKCat() function.

k <- chTKCat(
   host="localhost",                     # default parameter
   port=9111L,                           # default parameter
   drv=ClickHouseHTTP::ClickHouseHTTP(), # default parameter
   user="default",                       # default parameter
   password=""                           # if not provided the
                                         # password is requested interactively 
)

By default, this function connects anonymously (“default” user without password) to the database, using the HTTP interface of ClickHouse thanks to the ClickHouseHTTP driver. If the database is configured appropriately (see appendix), connection can be achieved through HTTPS with or without SSL peer verification (see the manual of ClickHouseHTTP::\ClickHouseHTTPDriver-class`for further information). Also, theRClickhouse::clickhouse()driver from the [RClickhouse][rclickhouse] package can be used (drvparameter of thechTKCat()` function) to leverage the native TCP interface of ClickHouse which has the strong advantage of having less overhead. But TLS wrapping is not supported yet by the RClickhouse package.

Once connected, this chTKCat object can be used as a TKCat object.

list_MDBs(k)             # get a specific MDBs from the catalog
## # A tibble: 2 × 11
##   name    title             descr…¹ url   version maint…² public popul…³ times…⁴
##   <chr>   <chr>             <chr>   <chr> <chr>   <chr>   <lgl>  <lgl>   <lgl>  
## 1 ClinVar Data extracted f… ClinVa… http… 0.13    Patric… TRUE   TRUE    TRUE   
## 2 HPO     Data extracted f… The Hu… http… 0.12    Patric… TRUE   TRUE    TRUE   
## # … with 2 more variables: timestamp <dttm>, access <fct>, and abbreviated
## #   variable names ¹​description, ²​maintainer, ³​populated, ⁴​timestamps
search_MDB_tables(k, "disease")  # Search table about "disease"
## # A tibble: 3 × 3
##   resource name                comment                 
##   <chr>    <chr>               <chr>                   
## 1 HPO      HPO_diseaseHP       HP presented by diseases
## 2 HPO      HPO_diseaseSynonyms Disease synonyms        
## 3 HPO      HPO_diseases        Diseases
search_MDB_fields(k, "disease")  # Search a field about "disease"
## # A tibble: 8 × 7
##   resource table               name    comment              type  nulla…¹ unique
##   <chr>    <chr>               <chr>   <chr>                <chr> <lgl>   <lgl> 
## 1 HPO      HPO_diseaseHP       db      Disease database     char… FALSE   FALSE 
## 2 HPO      HPO_diseaseSynonyms db      Disease database     char… FALSE   FALSE 
## 3 HPO      HPO_diseases        db      Disease database     char… FALSE   FALSE 
## 4 HPO      HPO_diseaseHP       id      Disease ID           char… FALSE   FALSE 
## 5 HPO      HPO_diseaseSynonyms id      Disease ID           char… FALSE   FALSE 
## 6 HPO      HPO_diseases        id      Disease ID           char… FALSE   FALSE 
## 7 HPO      HPO_diseases        label   Disease lable (pref… char… FALSE   FALSE 
## 8 HPO      HPO_diseaseSynonyms synonym Disease synonym      char… FALSE   FALSE 
## # … with abbreviated variable name ¹​nullable
collection_members(k)  
## # A tibble: 5 × 3
##   resource collection table              
##   <chr>    <chr>      <chr>              
## 1 ClinVar  BE         ClinVar_entrezNames
## 2 ClinVar  Condition  ClinVar_traitCref  
## 3 ClinVar  Condition  ClinVar_traits     
## 4 HPO      Condition  HPO_hp             
## 5 HPO      Condition  HPO_diseases
explore_MDBs(k)

4.3 Pushing an MDB in a chTKCat instance

Any MDB object can be imported in a TKCat ClickHouse instance as following:

kw <- chTKCat(host="localhost", port=9111L, user="pgodard")
create_chMDB(kw, "HPO", public=TRUE)
ch_hpo <- as_chMDB(file_hpo, kw)

It is then accessible to anyone with relevant permissions on the Clickhouse database. Pushing data in a ClickHouse database works only if the user is allowed to write in the database.

4.4 Specific operations on chMDB objects

The function get_MDB() returns a chMDB object that can be used as any MDB object. The data are located in the ClickHouse database and pulled on request.

ch_hpo <- get_MDB(k, "HPO")

To avoid pulling a whole table from ClickHouse (which can take time if the table is big), SQL queries can be made on the chMDB object as shown below.

get_query(
   ch_hpo,
   query="SELECT * from HPO_diseases WHERE lower(label) LIKE '%epilep%'"
)
## # A tibble: 265 × 3
##    db    id     label                                                           
##    <chr> <chr>  <chr>                                                           
##  1 OMIM  104130 Alopecia, psychomotor epilepsy, pyorrhea, and mental subnormali…
##  2 OMIM  117100 Centralopathic epilepsy                                         
##  3 OMIM  121201 Epilepsy, benign neonatal, 2                                    
##  4 OMIM  132090 Epilepsy, benign occipital                                      
##  5 OMIM  132300 Epilepsy, reading                                               
##  6 OMIM  159600 Myoclonic epilepsy, Hartung type                                
##  7 OMIM  159950 Spinal muscular atrophy with progressive myoclonic epilepsy     
##  8 OMIM  182610 Spastic paraplegia, epilepsy, and mental retardation            
##  9 OMIM  203600 Alopecia-Epilepsy-Oligophrenia syndrome of moynahan             
## 10 OMIM  208700 Ataxia with myoclonic epilepsy and presenile dementia           
## # … with 255 more rows

5 Appendices

5.1 chTKCat operations

5.1.1 Instantiating the ClickHouse database

5.1.1.1 Install ClickHouse, initialize and configure the TKCat instance

The ClickHouse docker container supporting TKCat, its initialization and its configuration procedures are implemented here: S01-install-and-init.R. This script should be adapted according to requirements and needs.

Specific attention should be paid on available ports: TCP native port (but not TLS wrapping yet) is supported by the RClickhouse R package whereas HTTP and HTTP ports are supported by the ClickHouseHTTP R package.

The data are stored in the TKCAT_HOME folder.

5.1.1.2 Cleaning and removing a TKCat instance

When no longer needed, stoping and removing the docker container can be achieved as exemplified below

# In shell
docker stop tkcat_test
docker rm tkcat_test
docker volume prune -f
# Remove the folder with all the data: `$TKCAT_HOME`.`
sudo rm -rf ~/Documents/Projects/TKCat_Test

5.1.2 User management

User management requires admin rights on the database.

5.1.2.1 Creation

k <- chTKCat(user="pgodard")
create_chTKCat_user(
   k, login="lfrancois", contact=NA, admin=FALSE, provider=TRUE
)

The function will require to setup a password for the new user. The admin parameter indicates if the new user have admin right on the whole chTKCat instance (default: FALSE). The provider parameter indicates if the new user can create and populate new databases whithin the chTKCat instance (default: FALSE).

5.1.2.2 Update

k <- chTKCat(user="pgodard")
change_chTKCat_password(k, "lfrancois")
update_chTKCat_user(k, contact="email", admin=FALSE)

A shiny application can be launched for updating user settings:

manage_chTKCat_users(k)

If this application is deployed, it can be made directly accessible from the explore_MDBs() Shiny application by providing the URL as the userManager parameter.

5.1.2.3 Drop

drop_chTKCat_user(k, login="lfrancois")

5.1.3 chMDB management

5.1.3.1 chMDB Creation

Before MDB data can be uploaded, the database should be created. This operation can only be achieved by data providers (see above).

create_chMDB(k, "CHEMBL", public=FALSE)

By default chMDB are not public. It can be changed through the public parameter when creating the chMDB or by using the set_chMDB_access() function afterward.

set_chMDB_access(k, "CHEMBL", public=TRUE)

Then, users having access to the chMDB can be identified with or without admin rights on the chMDB. Admin rights allow the user to update the chMDB data.

add_chMDB_user(k, "CHEMBL", "lfrancois", admin=TRUE)
# remove_chMDB_user(k, "CHEMBL", "lfrancois")
list_chMDB_users(k, "CHEMBL")

5.1.3.2 Populating chMDB

Each chMDB can be populated individualy using the as_chMDB() function. The code chunk below shows how to scan a directory for all fileMDB it contains. The as_memoMDB() function load all the data in memory and checks that all the model constraints are fulfilled (this step is optional). When overwrite parameter of the as_chMDB() function is set to FALSE (default), the potential existing version is archived before being updated. When overwrite is set to TRUE, the potential existing version is overwritten without being archived.

lc <- scan_fileMDBs("fileMDB_directory")
## The commented line below allows the exploration of the data models in lc.
# explore_MDBs(lc)
for(r in toFeed){
   message(r)
   lr <- as_memoMDB(lc[[r]])
   cr <- as_chMDB(lr, k, overwrite=FALSE)
}

5.1.3.3 Deleting a chMDB

Any admin user of a chMDB can delete the corresponding data.

empty_chMDB(k, "CHEMBL")

But only a system admin can drop the chMDB from the ClickHouse database.

drop_chMDB(k, "CHEMBL")

5.1.4 Collection management

Details about collections are provided in the following appendix.

Collections needs to be added to a chTKCat instance in order to support collection members of the different chMDB. They can be taken from the TKCat package environment, from a JSON file or directly from a JSON text variable. Additional functions are available to list and remove chTKCat collections.

add_chTKCat_collection(k, "BE")
list_chTKCat_collections(k)
remove_chTKCat_collection(k, "BE")

5.1.5 Implementation

5.1.5.1 Data models

5.1.5.1.1 Default database

The default database stores information about chTKCat instance, users and user access.

5.1.5.1.2 Modeled databases

Modeled databases (MDB) are stored in dedicated database in chTKCat. Their data model is provided in dedicated tables described below.

5.2 TKCat collections

Some MDBs refer to the same concepts and can be integrated accordingly. However they often use different vocabularies or scopes. Collections are used to identify such concepts and to define a way to document formally the scope used by the different members of these collections. Thanks to this formal description, tools can be used to automatically combine MDBs referring to the same collection but using different scopes, as shown above.

This appendix describes how to create TKCat Collections, document collection members and create functions to support the merging of MDBs.

5.2.1 Creating a collection

A collection is defined by a JSON document. This document should fulfill the requirements defined by the Collection-Schema.json. Two collections are available by default in the TKCat package.

list_local_collections()
## # A tibble: 2 × 2
##   title     description                                  
##   <chr>     <chr>                                        
## 1 BE        Collection of biological entity (BE) concepts
## 2 Condition Collection of condition concepts

Here is how the BE collection is defined.

get_local_collection("BE")
{
   "$schema": "https://json-schema.org/draft/2019-09/schema",
   "$id":"TKCat_BE_collection_1.0",
    "title": "BE collection",
    "type": "object",
    "description": "Collection of biological entity (BE) concepts",
    "properties": {
      "$schema": {"enum": ["TKCat_BE_collection_1.0"]},
      "$id": {"type": "string"},
        "collection": {"enum":["BE"]},
        "resource": {"type": "string"},
        "tables": {
            "type": "array",
            "minItems": 1,
            "items":{
                "type": "object",
                "properties":{
                    "name": {"type": "string"},
                    "fields": {
                        "type": "object",
                        "properties": {
                            "be": {
                                "type": "object",
                                "properties": {
                                    "static": {"type": "boolean"},
                                    "value": {"type": "string"}
                                },
                                "required": ["static", "value"],
                                "additionalProperties": false
                            },
                            "source": {
                                "type": "object",
                                "properties": {
                                    "static": {"type": "boolean"},
                                    "value": {"type": "string"}
                                },
                                "required": ["static", "value"],
                                "additionalProperties": false
                            },
                            "organism": {
                                "type": "object",
                                "properties": {
                                    "static": {"type": "boolean"},
                                    "value": {"type": "string"},
                                    "type": {"enum": ["Scientific name", "NCBI taxon identifier"]}
                                },
                                "required": ["static", "value", "type"],
                                "additionalProperties": false
                            },
                            "identifier": {
                                "type": "object",
                                "properties": {
                                    "static": {"type": "boolean"},
                                    "value": {"type": "string"}
                                },
                                "required": ["static", "value"],
                                "additionalProperties": false
                            }
                        },
                        "required": ["be", "source", "identifier"],
                        "additionalProperties": false
                    }
                },
                "required": ["name", "fields"],
                "additionalProperties": false
            }
        }
    },
    "required": ["$schema", "$id", "collection", "resource", "tables"],
    "additionalProperties": false
}

A collection should refer to the "TKCat_collections_1.0" $schema. It should then have the following properties:

  • $id: the identifier of the collection

  • title: the title of the collection

  • type: always object

  • description: a short description of the collection

  • properties: the properties that should be provided by collection members. In this case:

    • $schema: should be the $id of the collection

    • $id: the identifier of the collection member: a string

    • collection: should be “BE”

    • resource: the name of the resource having collection members: a string

    • tables: an array of tables corresponding to collection members. Each item being a table with the following features:

      • name: the name of the table

      • fields: the required fields

        • be: if static is true then value correspond to the be value valid for all the records. If not value correspond to the table column with the be value for each record.
        • source: if static is true then value correspond to the source value valid for all the records. If not value correspond to the table column with the source value for each record.
        • organism: if static is true then value correspond to the organism value valid for all the records. If not value correspond to the table column with the organism value for each record. type indicate how organisms are identified: "Scientific name" or "NCBI taxon identifier".

The main specifications defined in a JSON document can be simply displayed in R session by calling the show_collection_def() function.

get_local_collection("BE") %>%
   show_collection_def()
## BE collection: Collection of biological entity (BE) concepts
## Arguments (non-mandatory arguments are between parentheses):
##    - be:
##       + static: logical
##       + value: character
##    - source:
##       + static: logical
##       + value: character
##    - (organism):
##       + static: logical
##       + value: character
##       + type: character in 'Scientific name', 'NCBI taxon identifier'
##    - identifier:
##       + static: logical
##       + value: character

5.2.2 Documenting collection members

Documenting collection members of an MDB can be done by using the add_collection_member() function (as formerly described), or by writing a JSON file like the following one which correspond to BE members of the CHEMBL MDB.

system.file(
   "examples/CHEMBL/model/Collections/BE-CHEMBL_BE_1.0.json",
   package="TKCat"
) %>% 
   readLines() %>% paste(collapse="\n")
{
  "$schema": "TKCat_BE_collection_1.0",
  "$id": "CHEMBL_BE_1.0",
  "collection": "BE",
  "resource": "CHEMBL",
  "tables": [
    {
      "name": "CHEMBL_component_sequence",
      "fields": {
        "be": {
          "static": true,
          "value": "Peptide"
        },
        "identifier": {
          "static": false,
          "value": "accession"
        },
        "source": {
          "static": false,
          "value": "db_source"
        },
        "organism": {
          "static": false,
          "value": "organism",
          "type": "Scientific name"
        }
      }
    }
  ]
}

The identification of collection members should fulfill the requirements defined by the collection JSON document, and therefore pass the following validation.

jsonvalidate::json_validate(
   json=system.file(
      "examples/CHEMBL/model/Collections/BE-CHEMBL_BE_1.0.json",
      package="TKCat"
   ),
   schema=get_local_collection("BE"),
   engine="ajv"
)
## [1] TRUE

This validation is done automatically when reading a fileMDB object or when setting collection members with the add_collection_member() function.

5.2.3 Collection mapper functions

The merge.MDB() and the map_collection_members() functions rely on functions to map members of the same collection. When recorded (using the import_collection_mapper() function), these functions can be automatically identified by TKCat, otherwise or according to user needs, these functions could be provided using the funs (for merge.MDB()) or the fun (for map_collection_members()) parameters. Two mappers are pre-recorded in TKCat, one for the BE collection and one for the Condition collection. They can be retrieved with the get_collection_mapper() function.

get_collection_mapper("BE")
function (x, y, orthologs = FALSE, restricted = FALSE, ...) 
{
    if (!requireNamespace("BED")) {
        stop("The BED package is required")
    }
    if (!BED::checkBedConn()) {
        stop("You need to connect to a BED database using", " the BED::connectToBed() function")
    }
    if (!"organism" %in% colnames(x)) {
        d <- x
        scopes <- dplyr::distinct(d, be, source)
        nd <- c()
        for (i in 1:nrow(scopes)) {
            be <- scopes$be[i]
            source <- scopes$source[i]
            toadd <- d %>% dplyr::filter(be == be, source == 
                source)
            organism <- BED::guessIdScope(toadd$identifier, be = be, 
                source = source, tcLim = Inf) %>% attr("details") %>% 
                filter(be == !!be & source == !!source) %>% pull(organism) %>% 
                unique()
            toadd <- merge(toadd, tibble(organism = organism))
            nd <- bind_rows(nd, toadd)
        }
        x <- nd %>% mutate(organism_type = "Scientific name")
    }
    if (!"organism" %in% colnames(y)) {
        d <- y
        scopes <- dplyr::distinct(d, be, source)
        nd <- c()
        for (i in 1:nrow(scopes)) {
            be <- scopes$be[i]
            source <- scopes$source[i]
            toadd <- d %>% dplyr::filter(be == be, source == 
                source)
            organism <- BED::guessIdScope(toadd$identifier, be = be, 
                source = source, tcLim = Inf) %>% attr("details") %>% 
                filter(be == !!be & source == !!source) %>% pull(organism) %>% 
                unique()
            toadd <- merge(toadd, tibble(organism = organism))
            nd <- bind_rows(nd, toadd)
        }
        y <- nd %>% mutate(organism_type = "Scientific name")
    }
    xscopes <- dplyr::distinct(x, be, source, organism, organism_type)
    yscopes <- dplyr::distinct(y, be, source, organism, organism_type)
    toRet <- NULL
    for (i in 1:nrow(xscopes)) {
        xscope <- xscopes[i, ]
        if (any(apply(xscope, 2, is.na))) {
            (next)()
        }
        xi <- dplyr::right_join(x, xscope, by = c("be", "source", 
            "organism", "organism_type"))
        xorg <- ifelse(xscope$organism_type == "NCBI taxon identifier", 
            BED::getOrgNames(xscope$organism) %>% dplyr::filter(nameClass == 
                "scientific name") %>% dplyr::pull(name), xscope$organism)
        for (j in 1:nrow(yscopes)) {
            yscope <- yscopes[j, ]
            if (any(apply(yscope, 2, is.na))) {
                (next)()
            }
            yi <- dplyr::right_join(y, yscope, by = c("be", "source", 
                "organism", "organism_type"))
            yorg <- ifelse(yscope$organism_type == "NCBI taxon identifier", 
                BED::getOrgNames(yscope$organism) %>% dplyr::filter(nameClass == 
                  "scientific name") %>% dplyr::pull(name), yscope$organism)
            if (xorg == yorg || orthologs) {
                xy <- BED::convBeIds(ids = xi$identifier, from = xscope$be, 
                  from.source = xscope$source, from.org = xorg, 
                  to = yscope$be, to.source = yscope$source, 
                  to.org = yorg, restricted = restricted) %>% 
                  dplyr::as_tibble() %>% dplyr::select(from, 
                  to)
                if (restricted) {
                  xy <- dplyr::bind_rows(xy, BED::convBeIds(ids = yi$identifier, 
                    from = yscope$be, from.source = yscope$source, 
                    from.org = yorg, to = xscope$be, to.source = xscope$source, 
                    to.org = xorg, restricted = restricted) %>% 
                    dplyr::as_tibble() %>% dplyr::select(to = from, 
                    from = to))
                }
                xy <- xy %>% dplyr::rename(identifier_x = "from", 
                  identifier_y = "to") %>% dplyr::mutate(be_x = xscope$be, 
                  source_x = xscope$source, organism_x = xscope$organism, 
                  be_y = yscope$be, source_y = yscope$source, 
                  organism_y = yscope$organism)
                toRet <- dplyr::bind_rows(toRet, xy)
            }
        }
    }
    toRet <- dplyr::distinct(toRet)
    return(toRet)
}

A mapper function must have at least an x and a y parameters. Each of them should be a data.frame with all the field values corresponding to the fields defined in the collection. Additional parameters can be defined and will be forwarded using .... This function should return a data frame with all the fields values followed by “_x” and “_y” suffix accordingly.

5.3 Remarks about supported data format and data types

Most of the data format and data types supported by the ReDaMoR and the TKCat packages are taken into account in the examples described in the main sections of this vignette. Nevertheless, one specific data format (matrix) and one specific data type (base64) are not exemplified. This appendix provides a short description of these format and type.

5.3.1 Matrices of values

ReDaMoR and TKCat support data frame and matrix objectq. Data frame is the most used data format from far. However, matrices of values can be useful in some use cases. The example below shows how such data format are modeled in ReDaMoR as a 3 columns table: one of type “row” corresponding to the row names of the matrix, one of type “column” corresponding to the column names of the matrix, and one of any type (excepted “row”, “column”, or “base64”).

d <- matrix(
   rnorm(40), nrow=10,
   dimnames=list(
      paste0("g", 1:10),
      paste0("s", 1:4)
   )
)
m <- ReDaMoR::df_to_model(d) %>% 
   ReDaMoR::rename_field("d", "row", "gene") %>%
   update_field("d", "gene", comment="Gene identifier") %>% 
   ReDaMoR::rename_field("d", "column", "sample") %>% 
   update_field("d", "sample", comment="Sample identifier") %>% 
   ReDaMoR::rename_field("d", "value", "expression") %>% 
   update_field(
      "d", "expression", nullable=FALSE, comment="Gene expression value"
   )
md <- memoMDB(list(d=d), m, list(name="Matrix example"))
plot(data_model(md))

5.3.2 Documents stored as base64 values

Whole documents can be stored in MDB as “base64” character values. The example below shows how a document can be put in a table and the corresponding data model.

ch_config_files <- tibble(
   name=c("config.xml", "users.xml"),
   file=c(
      base64enc::base64encode(
         system.file("ClickHouse/config.xml", package="TKCat")
      ),
      base64enc::base64encode(
         system.file("ClickHouse/users.xml", package="TKCat")
      )
   )
)
m <- df_to_model(ch_config_files) %>% 
   update_field(
      "ch_config_files", "name",
      type="base64", comment="Name of the config file",
      nullable=FALSE, unique=TRUE
   ) %>% 
   update_field(
      "ch_config_files", "file",
      type="base64", comment="Config file in base64 format",
      nullable=FALSE
   )
md <- memoMDB(
   list(ch_config_files=ch_config_files), m, list(name="base64 example")
)
plot(data_model(md))

Acknowledgments

This work was entirely supported by UCB Pharma (Early Solutions department).

References

François, Liesbeth, Jonathan van Eyll, and Patrice Godard. 2020. “Dictionary of Disease Ontologies (DODO): A Graph Database to Facilitate Access and Interaction with Disease and Phenotype Ontologies.” F1000Research 9 (August): 942. https://doi.org/10.12688/f1000research.25144.1.
Godard, Patrice, and Jonathan van Eyll. 2018. BED: A Biological Entity Dictionary Based on a Graph Data Model.” F1000Research 7: 195. https://doi.org/10.12688/f1000research.13925.3.
Kinsella, R. J., A. Kahari, S. Haider, J. Zamora, G. Proctor, G. Spudich, J. Almeida-King, et al. 2011. “Ensembl BioMarts: A Hub for Data Retrieval Across Taxonomic Space.” Database 2011 (0): bar030–30. https://doi.org/10.1093/database/bar030.
Köhler, Sebastian, Leigh Carmody, Nicole Vasilevsky, Julius O B Jacobsen, Daniel Danis, Jean-Philippe Gourdine, Michael Gargano, et al. 2019. “Expansion of the Human Phenotype Ontology (HPO) Knowledge Base and Resources.” Nucleic Acids Research 47 (D1): D1018–27. https://doi.org/10.1093/nar/gky1105.
Landrum, Melissa J., Jennifer M. Lee, Mark Benson, Garth R. Brown, Chen Chao, Shanmuga Chitipiralla, Baoshan Gu, et al. 2018. ClinVar: Improving Access to Variant Interpretations and Supporting Evidence.” Nucleic Acids Research 46 (D1): D1062–67. https://doi.org/10.1093/nar/gkx1153.
Mendez, David, Anna Gaulton, A Patrícia Bento, Jon Chambers, Marleen De Veij, Eloy Félix, María Paula Magariños, et al. 2019. ChEMBL: Towards Direct Deposition of Bioassay Data.” Nucleic Acids Research 47 (D1): D930–40. https://doi.org/10.1093/nar/gky1075.
Wu, Chunlei, Ian MacLeod, and Andrew I. Su. 2012. “BioGPS and MyGene.info: Organizing Online, Gene-Centric Information.” Nucleic Acids Research 41 (D1): D561–65. https://doi.org/10.1093/nar/gks1114.