ENCODExplorer: A compilation of metadata from ENCODE

In ENCODExplorer: A compilation of ENCODE metadata

This package and the underlying code are distributed under the Artistic license 2.0. You are free to use and redistribute this software.

Rationale

"The ENCODE (Encyclopedia of DNA Elements) Consortium is an international collaboration of research groups funded by the National Human Genome Research Institute (NHGRI). The goal of ENCODE is to build a comprehensive parts list of functional elements in the human genome, including elements that act at the protein and RNA levels, and regulatory elements that control cells and circumstances in which a gene is active" source: ENCODE Projet Portal.

However, retrieving and downloading data can be time consuming using the current web portal, especially when multiple files from different experiments are involved.

This package has been designed to facilitate access to ENCODE data by compiling the metadata associated with files, experiments, datasets, biosamples, and treatments.

We implemented time-saving features to select ENCODE files by querying their metadata, downloading them and validating that the file was correctly downloaded.

This vignette will introduce the main features of the ENCODExplorer package.

Loading the ENCODExplorer package

library(ENCODExplorer)

Introduction

To use the functionalities of the ENCODExplorer package, you must first download the data.table containing all of the ENCODE metadata.

This data.table is available through the AnnotationHub package. For convenience, the latest available version at each release will be downloaded and used by default.

We also provide the following function to quickly obtain these metadata:

To load getencode_df :

encode_df <- get_encode_df()

Main functions

Query

The queryEncode function allows the user to find the subset of files corresponding to a precise query defined according to the following criteria :

|Parameter| Description| |---------|-----------------------------------------| |set_accession|The accession for the containing experiment or dataset| |dataset_accession|There is a subtle difference between the parameters set_accession and dataset_accession. In fact, some files can be part of an experiment, a dataset or both. When using set_accession, you will get all the files directly associated with this accession (experiment and/or dataset). While the usage of dataset_accession will get the files directly associated to the requested dataset AND those which are part of an experiment and indirectly linked to a dataset (reported as related files in the dataset and related_dataset in the experiment).| |file_accession|The accesion for one specific file| |biosample_name|The biosample name ("GM12878", "kidney")| |biosample_type|The biosample type ("tissue", "cell line")| |assay|The assay type ("ChIP-seq", "polyA plus RNA-seq")| |file_format|The file format. Some currently available formats include bam, bed, fastq, bigBed, bigWig, CEL, csfasta, csqual, fasta, gff, gtf, idat, rcc, sam, tagAlign, tar, tsv, vcf, wig.| |lab|The laboratory| |organism|The donor organism ("Homo sapiens", "Mus musculus")| |target|The gene, protein or histone mark which was targeted by the assay (Immunoprecipitated protein in ChIP-seq, knocked-down gene in CRISPR RNA-seq assays, etc)| |treatment|The treatment related to the biosample| |project|The project name/id|

By default, the query function uses exact string matching to perform the selection of the relevant entries. This behavior can be changed by modifying the fixed or fuzzy parameters. Setting fixed to FALSE will perform case-insensitive regular expression matching. Setting fuzzy to TRUE will retrieve search results where the query string is a partial match.

The result set is a subset of the encode_df_lite table.

For example, to select all fastq files originating from assays on the MCF-7 (human breast cancer) cell line:

query_results <- queryEncode(organism = "Homo sapiens", 
                      biosample_name = "MCF-7", file_format = "fastq",
                      fixed = TRUE)

The same request with approximate spelling of the biosample name will return no results:

query_results <- queryEncode(organism = "Homo sapiens", biosample_name = "mcf7",
                        file_format = "fastq", fixed = TRUE,
                        fuzzy = FALSE)

However, if you follow the warning guidance and set the fuzzy parameter to TRUE:

query_results <- queryEncode(organism = "Homo sapiens",
                    biosample_name = "mcf7", file_format = "fastq",
                    fixed = TRUE, fuzzy = TRUE)

You can also perform matching through regular expressions by setting fixed to FALSE.

query_results <- queryEncode(assay = ".*RNA-seq",
                    biosample_name = "HeLa-S3", fixed = FALSE)
table(query_results$assay)

Finally, the queryEncodeGeneric function can be used to perform searches on columns which are not part of the queryEncode interface but are present within the encode_df_lite data.table:

query_results <- queryEncodeGeneric(biosample_name="HeLa-S3",
                    assay="RNA-seq", submitted_by="Diane Trout",
                    fuzzy=TRUE)
table(query_results$submitted_by)

These criteria correspond to the filters that you can find on ENCODE portal:

fuzzySearch

This function is a more user-friendly version of queryEncode that also perform searches on the encode_df_lite object. The character vector or the list of characters specified by the user will be searched for in every column of the database. The user can also constrain the query by selecting the specific columns in which to search for the query term by using the filterVector parameter.

The following request will produce a data.table with every files containing the term brca.

fuzzy_results <- fuzzySearch(searchTerm = c("brca"))

Multiple terms can be searched simultaneously. This example extracts all files containing brca or ZNF24 within the target column.

fuzzy_results <- fuzzySearch(searchTerm = c("brca", "ZNF24"),
                             filterVector = c("target"),
                             multipleTerm = TRUE)

When searching for multiple terms, three type of input can be passed to the searchTerm parameter : - A single character where the various terms are separated by commas - A character vector - A list of characters

Search

This function simulates a keyword search performed through the ENCODE web portal.

The searchEncode function returns a data frame corresponding to the result page provided by the ENCODE portal. If a specific file or dataset isn't available with fuzzySearch or queryEncode (i.e. within get_encode_df()), the user can access the latest data from the ENCODE database through the searchEncode function.

The searchToquery function convert the result of a search to a data.table with the same design as get_encode_df(). This format contains more metadata and allow the user to extract all files within the dataset. This format also allows the user to create a design using the createDesign function.

Here is the example of the following search : "a549 chip-seq homo sapiens".

On ENCODE portal :

With our function :

  search_results <- searchEncode(searchTerm = "a549 chip-seq homo sapiens",
                                 limit = "all")

createDesign

This function organizes the data.table created by fuzzySearch, queryEncode or searchToquery. It extracts the replicate and control files within a dataset.

It creates a data.table with the file accessions, the dataset accessions and numeric values associated with the nature of the file (1:replicate / 2:control) when the format parameter is set to long.

By setting the format parameter to wide, each dataset will have its own column as illustrated below.

downloadEncode

downloadEncode allows a user to download a file or an entire dataset. Downloading files can be done by providing a vector of file accessions or dataset accessions (represented by the accession column in get_encode_df()) to the file_acc parameter. This parameter can also be the data.table created by queryEncode, fuzzySearch, searchToquery or createDesign.

If the accession doesn't exist within the passed-in get_encode_df() database, downloadEncode will search for the accession directly within the ENCODE database. The path to the download directory can be specified (default: /tmp).

To ensure the integrity of each file, the md5 sum of each downloaded file is compared to the reported md5 sum in ENCODE.

Moreover, if the accession is a dataset accession, the function will download each file in this dataset. The format option, which is set by default to all, enables the downloading of a specific format.

Here is a small example query:

query_results <- queryEncode(assay = "switchgear", target ="elavl1", fixed = FALSE)

And its equivalent search:

search_results <- searchEncode(searchTerm = "switchgear elavl1", limit = "all")

To select a particular file format you can:

1) add filters to your query and then run the downloadEncode function.

query_results <- queryEncode(assay = "switchgear", target ="elavl1",
                             file_format = "bed" , fixed = FALSE)
downloadEncode(query_results)

2) specify the format to the downloadEncode function.

downloadEncode(search_results, format = "bed")

Conversion

The function searchToquery enables the conversion of the results of searchEncode to a queryEncode output based on the accession numbers. The user can then benefit from all the collected metadata and the createDesign function.

The structure of the result set is similar to the get_encode_df() structure.

Let's try it with the previous example :

1) search

search_results <- searchEncode(searchTerm = "switchgear elavl1", limit = "all")

2) convert

convert_results <- searchToquery(searchResults = search_results)

shinyEncode

This function launches the shinyApp of ENCODExplorer that implements the fuzzySearch and queryEncode search functions. It also allows the creation of a design to organize and download specific files with the downloadEncode function. The Search tab of shinyEncode uses the fuzzySearch function for a low specificity request while the Advanced Search tab uses the queryEncode function.

Summarizing ENCODE data

While queryEncode, searchEncode and downloadEncode gives the user access to ENCODE's raw files, ENCODExplorer also provides helper functions which load and summarize ENCODE data for common biological questions.

Obtaining consensus peaks from ChIP-Seq

The most common question in a ChIP-Seq assay is: "Where does the protein of interest bind the genome?" To answer this question, ENCODExplorer provides the queryConsensusPeaks method. queryConsensusPeaks finds all ChIP-seq peak files matching the given criteria, split them by treatment group, and builds a set of "consensus peaks". The "consensus peaks" identified by ENCODExplorer are those that appear in all replicates of a given group. Two peaks are considered to belong to the same binding event if at least one of their nucleotides overlap.

# Obtain a summary of all peaks for CTCF ChIP-Seq assays in the 22Rv1
# (human prostate carcinoma) cell line.
res = queryConsensusPeaks("22Rv1", "GRCh38", "CTCF")

The list of downloaded files is available through the files() method.

files(res)

The metadata for those file is available through the file_metadata() method. The file metadata are split according to treatment group:

f_meta = file_metadata(res)
names(f_meta)
f_meta[[1]][,1:5]

A data-frame explaining how each treatment group was split is available through the metadata() method:

metadata(res)

The list of all peaks identified in individual files are accessed through the peaks() method.

names(peaks(res))
names(peaks(res)[[1]])
peaks(res)[[1]]

Finally, the consensus peaks (those who are present in all individual replicates) are accessed through the consensus() method:

names(consensus(res))
consensus(res)

Fine-tuning a consensus peaks query

Certain versions of the ENCODE pipeline provide multiple calling algorithms. Also, sometimes multiple labs have performed ChIP-seq experiments on the same tissue and protein, and these results might not be directly comparable. ENCODExplorer uses heuristics to try and determine which set of files will provide the most informative results, but the results of these heuristics might prove unsatisfactory.

In such cases, a user can provide his own set of ENCODE metadata and his own choice of splitting columns using the buildQueryConsensus function. The user can also specify which proportion of individual replicates a peak must appear in to be included in the consensus peaks through the consensus_threshold parameter:

query_results = queryEncodeGeneric(biosample_name="A549", assembly="GRCh38",
                                   file_format="^bed$", output_type="^peaks$", 
                                   treatment_duration_unit="minute",
                                   treatment_duration="(^5$|^10$)", 
                                   target="NR3C1", fixed=FALSE)

# Obtain a summary of all peaks for NR3C1 ChIP-Seq assays in the A549
# cell line.
res = buildConsensusPeaks(query_results, split_by=c("treatment_duration"), 
                          consensus_threshold=0.5)

res

Obtaining average gene expression

For RNA-Seq experiment, the most straightforward type of results is the expression level of all genes or transcripts. ENCODExplorer provides the queryGeneExpression and queryTranscriptExpression methods to summarize these results. ENCODExplorer finds all gene or transcript expression levels for a given biosample and calculates per-condition mean values.

Most biosamples in the ENCODE Project have RNA-seq experiments targeting different cell fractions, such as whole cells, cytoplasmic fractions, and nuclear fractions. Since it makes no biological sense to aggregate such results, ENCODExplorer automatically splits them by the dataset_description column, which details the cell fraction as well as other methodological or biological parameters which make samples unfit for aggregation.

# Obtain a summary of all peaks for NR3C1 ChIP-Seq assays in the A549
# cell line.
res = queryGeneExpression("bone marrow")

The files(), file_metadata() and metadata() methods behave the same way as they do for queryConsensusPeaks:

metadata(res)

You can see which expression metric ENCODExplorer extracted using the metric() method.

metric(res)

Per gene/transcript values for all metrics are available through the metric_data() method:

head(metric_data(res))

You can also get a list of the raw ENCODE files by calling the raw_data() method.

head(raw_data(res)[[1]][[1]])

Fine tuning expression summaries

Just as it is the case for ChIP-seq assays, it can sometimes be easier for the user to perform filtering of the ENCODE results manually. For thse cases, ENCODExplorer provides the buildExpressionSummary method.

query_results = queryEncodeGeneric(biosample_name="neural tube", 
                                   output_type="gene quantifications",
                                   file_type="tsv",
                                   assay="polyA plus RNA-seq",
                                   assembly="^mm10$",
                                   dataset_biosample_summary="(15.5|13.5)",
                                   fixed=FALSE)

buildExpressionSummary(query_results, split_by="dataset_biosample_summary")                                   

Interactive mode

queryConsensusPeaks, queryGeneExpression and queryTranscriptExpression all make educated guesses about the assembly, assay and sample types to be used for generating the summaries. However, by setting the use_interactive argument to TRUE, a user can take direct control of some these choices.

queryGeneExpression("neural tube", use_interactive=TRUE)

Updating the ENCODE file database

By default, ENCODExplorer retrieves the ENCODE metadata from its sister package, r Biocpkg("ENCODExplorerData"). The version of the metadata provided by default will be updated with each Bioconductor release in the ENCODExplorer package. However, since all of ENCODExplorer's function take an explicit df parameter, it is possible to use the r Biocpkg("AnnotationHub") package to download a more recent version:

require(AnnotationHub)
ah = AnnotationHub()
query(ah, "ENCODExplorerData")

Finally, it is also possible to use r Biocpkg("ENCODExplorerData") functionalities to generate an up-to-date data.table, and pass it to ENCODExplorer's functions.

We refer the user to the r Biocpkg("ENCODExplorerData") vignettes for details on how to generate an up-to-date data.table.

ENCODExplorer documentation built on Nov. 8, 2020, 5:11 p.m.