FAIRification of an RNAseq dataset

Overview
Creative Commons License: CC-BY Questions:
  • How can an RNAseq dataset be made FAIR?

  • How are the FAIR principles put into practice with a data-type used commonly in the Life Sciences?

Objectives:
  • To be able to map each of the FAIR principles to a dataset in the public domain

Requirements:
Time estimation: 1 hour
Supporting Materials:
Published: Mar 27, 2024
Last modification: Mar 27, 2024
License: Tutorial Content is licensed under Creative Commons Attribution 4.0 International License. The GTN Framework is licensed under MIT
purl PURL: https://gxy.io/GTN:T00435
version Revision: 1

RNA sequencing is chosen here as an example of how to FAIRify data for a popular assay in the Life Sciences. RNAseq data can be shared and curated in designated public repositories using established ontologies (and controlled vocabularies) for describing protocols and biological material (metadata).

Two international repositories are commonly used to locate and download RNAseq (meta)data: ArrayExpress and GEO. Other repositories for raw sequence data exist (e.g. SRA, ENA, DDBJ), but ArrayExpress and GEO specifically house and index expression data , including rich metadata detailing samples, data processing and final results files such as gene expression matrices.

By submitting data to a public repository, it becomes openly accessible, searchable and annotated with rich metadata, by the submitter and curation team. Note, both repositories belong to the FAIRsharing database registry, which can help you find public repositories for all types of Life Science data.

This lesson will take you through a publicly available RNAseq dataset in ArrayExpress and show you how it meets FAIR principles using the checklist published in 2016 Wilkinson et al. 2016.

Agenda

In this tutorial, we will cover:

  1. Example of FAIRification when data is uploaded to a public repository
  2. Finding and accessing an RNAseq dataset
    1. (F1) (Meta)data are assigned a globally unique and persistent identifier
    2. (A1) (Meta)data are retrievable by their identifier using a standardised communications protocol (A1.1) The protocol is open, free, and universally implementable
    3. (F4) (Meta)data are registered or indexed in a searchable resource
  3. Learning about the experiment and dataset from reading the metadata
    1. (F1) (Meta)data are assigned a globally unique and persistent identifier (F3) Metadata clearly and explicitly include the identifier of the data they describe
    2. (F2) Data are described with rich metadata
  4. Metadata and community standards
    1. (I1) (Meta)data use a formal, accessible, shared, and broadly applicable language for knowledge representation.
    2. (I2) (Meta)data use vocabularies that follow FAIR principles. (R1) (Meta)data are richly described with a plurality of accurate and relevant Attributes.
    3. (R1.3) (Meta)data meet domain-relevant community standards.
  5. Downloading data and metadata for reuse
    1. (R1.1) (Meta)data are released with a clear and accessible data usage license
    2. (R1.2) (Meta)data are associated with detailed provenance.

Example of FAIRification when data is uploaded to a public repository

We will use the following human RNAseq dataset for this learning material ArrayExpress:E-MTAB-8316. This link hosts all metadata and links to downloadable raw and transformed data, shown as a schematic below.

A schematic of RNAseq data and metadata within ArrayExpress. Open image in new tab

Figure 1: A schematic of RNAseq data and metadata within ArrayExpress. Data shown are raw FASTQ files and summary gene transcript count matrices.

Finding and accessing an RNAseq dataset

In this section, four of the FAIR Guiding Principles are put into practice:

  • (F1) (Meta)data are assigned a globally unique and persistent identifier
  • (F4) (Meta)data are registered or indexed in a searchable resource
  • (A1) (Meta)data are retrievable by their identifier using a standardised communications protocol
  • (A1.1) The protocol is open, free, and universally implementable

(F1) (Meta)data are assigned a globally unique and persistent identifier

Question

The RNAseq dataset we are looking at belongs to a study published in Nature Medicine Alivernini et al. 2020. Can you find the globally unique and persistent identifier for the RNAseq data and data descriptions (metadata), within the text of the publication?

Under “Data and code availability’, towards the end of the manuscript, we are told the following.

The data availability section within the RNAseq publication. Open image in new tab

Figure 2: The data availability section within the RNAseq publication showing links to access primary data.

Note that the dataset’s unique and persistent (unchanging) identifier is E-MTAB-8316, (F1) and could also be expressed as the full [URL](https://www.ebi.ac.uk/biostudies/ArrayExpress/studies/E-MTAB-8316. Note that there are two other identifiers given to 2 single cell RNAseq datasets published as part of the same study. For this lesson, we are only considering the bulk RNAseq dataset.

Note also, the same record can be accessed using a resolution service, such as identifiers.org, that allows URLs to be regularised using a namespace (arrayexpress) and the identifier local to the database (E-MTAB-8316)

(A1) (Meta)data are retrievable by their identifier using a standardised communications protocol (A1.1) The protocol is open, free, and universally implementable

Question

Use the following URL to access this dataset. The URL, in this case, is a concatenation of the URL of the data repository and the unique identifier:

You should see the following webpage hosted at the EBI (www.ebi.ac.uk). In the context of FAIR, you have now accessed this RNAseq data using a weblink employing https, where https is a standardised communications protocol or “data getting method” that is open, free and universally implementable (A1, A1.1). Note that the title of this RNAseq record and the title of the companion publication are different, which means the two cannot be linked by title. Instead, the accession (persistent ID) is used to connect the data and companion publication.

Many databases, including ArrayExpress, also have programmatic ways to access and download data. For advanced users, ArrayExpress host a REST server for this purpose. Other packages written by third parties, such as ffq, also permit this function.

RNAseq data record. Open image in new tab

Figure 3: The RNAseq data record showing the Primary ID circled in red.

(F4) (Meta)data are registered or indexed in a searchable resource

Question

Location the dataset using the persistent ID in a different way. Use the search menu in ArrayExpress to find the dataset again. Use the following link to (access ArrayExpress)[https://www.ebi.ac.uk/biostudies/arrayexpress/studies/] and then type E-MTAB-8316 into the search bar.

Note here that data access is gained through searching the database whereas previously we used a direct URL. We are able to search ArrayExpress with the persistent ID because the RNAseq dataset and its metadata are indexed (F4).

ArrayExpress/BioStudies search bar. Open image in new tab

Figure 4: The ArrayExpress/BioStudies search bar with the query circled in red.
Question

Alternatively, use the same search menu in ArrayExpress to search again using the words “macrophage rheumatoid arthritis” and selecting “rna-seq of coding rna” on the left-hand search bar.

Here, the dataset we want is not the first in the list but appears later in the search results. In this example, data access is gained through searching metadata (data about the experiment) (F4), and not the persistent ID as we did previously.

ArrayExpress/BioStudies search bar for the metadata query. Open image in new tab

Figure 5: The ArrayExpress/BioStudies search bar showing a metadata query circled in red.

Learning about the experiment and dataset from reading the metadata

In this section, the Findability Principles are put into practice:

  • (F1) (Meta)data are assigned a globally unique and persistent identifier
  • (F2) Data are described with rich metadata
  • (F3) Metadata clearly and explicitly include the identifier of the data they describe

(F1) (Meta)data are assigned a globally unique and persistent identifier (F3) Metadata clearly and explicitly include the identifier of the data they describe

Question

Start looking at the metadata given on dataset webpage. Try to find the unique, persistent identifier in the record (F1) (Meta)data are assigned a globally unique and persistent identifier.

The persistent identifier is circled in red and is the first thing we see in the record (F1). All metadata (descriptions about the data) and the actual raw data files are linked from this page. Here metadata clearly and explicitly include the identifier of the data they describe (F3).

RNAseq data record showing the Primary ID (persistent identifier). Open image in new tab

Figure 6: The RNAseq data record showing the Primary ID (persistent identifier) circled in red.

Here data are described with rich metadata (F2), which allows a person to reuse data appropriately by reducing ambiguity around what the data mean or how they are derived. This principle applies also to other principles in FAIR such as (R1) Meta(data) are richly described with a plurality of accurate and relevant attributes.

(F2) Data are described with rich metadata

Indexing rich metadata allows a person to easily locate a dataset of interest, since it is made searchable within ArrayExpress. Metadata is added by the person submitting the data and is further curated by the ArrayExpress database. Metadata curation is performed via a web-based submission interface called Annotare which aids rich curation through using community ontologies and controlled vocabularies.

Question

Familiarise yourself with the page layout: there are links to all protocols, data, sample metadata and assay type. How many samples are in this dataset?

There are 12 samples in this dataset.

The RNAseq data record showing the number of samples (assays). Open image in new tab

Figure 7: The RNAseq data record showing the number of samples (assays) circled in red.
Question

What data provenance can you find? i.e. what processes have been performed to create the data linked from this page? Are there any metadata which might be absent?

The final 2 protocols within the protocol table drawdown detail all data transformations for the raw and transformed data.

The protocol table from the RNAseq data record. Open image in new tab

Figure 8: The protocol table from the RNAseq data record showing the final 2 data protocols circled in red.

When recording data provenance, ambiguities can be introduced. True provenance can be achieved though through publishing scripts and software versions, enabling other researchers to repeat all data transformation processes.

Metadata and community standards

In this section, we will consider Interoperability and data Reuse:

  • (I1) (Meta)data use a formal, accessible, shared, and broadly applicable language for knowledge representation.
  • (I2) (Meta)data use vocabularies that follow FAIR principles.
  • (R1) (Meta)data are richly described with a plurality of accurate and relevant Attributes.
  • (R1.3) (Meta)data meet domain-relevant community standards.

Interoperability is mentioned in our companion (FAIR Pointers course)[https://fellowship.elixiruknode.org/latest/carpentries-course-fair-pointers] where we consider merging datasets using a common controlled vocabulary.

(I1) (Meta)data use a formal, accessible, shared, and broadly applicable language for knowledge representation.

Commonly when adding metadata to a dataset a published ontology is used. In its simplest form, an ontology is a dictionary of terms you use to annotate a piece of data. For example, the NCBI taxonomy database is something you may have used to annotate species within a dataset. An ontology, though, will also define relationships between terms. So in the taxonomy example, the term “Homo sapiens” will belong to parent terms such as Primate, Mammal and so on. Importantly, by using an ontology you can ensure you are using interoperable and searchable terms with your data.

(I2) (Meta)data use vocabularies that follow FAIR principles. (R1) (Meta)data are richly described with a plurality of accurate and relevant Attributes.

Published ontologies are linked from the ArrayExpress data submission tool, Annotare, so this work is done for you. Although, if you wish to start using ontologies to annotate your data at the point it is produced, the FAIRsharing Standards registry is a place to start. Importantly, when an ontology is published it too becomes findable and reusable in the context of the FAIR principles (I2). Metadata using published ontologies permit interoperability since you can match identical annotations across data and databases (I1 & I2). Additionally, they are created, refined and used by communities of practice (R1, R1.3).

Question

Look at the page again. Identify any metadata that belongs to an existing, published ontology. Note, we have mentioned one already: “Homo sapiens” as part of taxonomy.

There are many. There are ontologies used for Source Characteristics such as ‘Developmental stage’, ‘Disease’, ‘Organism part’, ‘Cell type’, and so on. These annotations are curated at the point of data submission.

published ontologies metadata. Open image in new tab

Figure 9: Metadata within the record using controlled terms from published ontologies.

There are also others that are not so obvious. Under “Protocols’, a protocol ontology is used under the column ‘Type’. By selecting the EFO link-and-arrow following the protocol annotation, you are taken to the FAIR (Experimental factor ontology)[https://www.ebi.ac.uk/ols/ontologies/efo] at the EBI.

Experimental Factor Ontology protocol table from the RNAseq data record. Open image in new tab

Figure 10: The protocol table from the RNAseq data record showing the final 2 data protocols circled in red with links to the Experimental Factor Ontology.

(R1.3) (Meta)data meet domain-relevant community standards.

fair RNAseq MINSEQ. Open image in new tab

Figure 11: The MINSEQ Score table from the RNAseq data.

Notice that the “MINSEQE” star rating on the left hand banner, indicates a level of Community Compliance in terms of (meta)data submission. The community standard in this case is MINSEQE (Minimum Information about a Sequencing Experiment Checklist) which is recorded in FAIRsharing Brazma et al. 2012. This example receives a maximum 5-star score demonstrating that data and metadata are available for experiential design, protocols and variables as well as processed and raw data.

At a more advanced level of community compliance, the ArrayExpress database uses a domain specific metadata format: MAGE-TAB, also identified in FAIRsharing. MAGE-Tab is a tab-delimited file containing all relevant metadata for the experiment. This is a machine-readable version of the webpage we are looking at.

Hands-on: Locate the MAGE-TAB file containing sample metadata, and open in Excel.

Notice how all data on the webpage are also contained in this machine-readable format. You are looking to download the SDRF file (Sample and Data Relationship Format).

fair RNAseq links. Open image in new tab

Figure 12: Links to machine-readable metadata in the RNAseq data record circled in red.

Finally, this ArrayExpress record links to the raw sequence data housed in the ENA (European Nucleotide archive). We see this in an exercise later in this course. The ENA record in turn links to ENA SRA XML format (community recognised), which gives another machine-readable interoperability format.

Hands-on: Machine Readable RNA XML formats

Have a quick look at the machine readable RNA XML formats. Notice how data are contained within tags very similar to HTML.

Downloading data and metadata for reuse

  • (R1.1) (Meta)data are released with a clear and accessible data usage license.
  • (R1.2) (Meta)data are associated with detailed provenance.

(R1.1) (Meta)data are released with a clear and accessible data usage license

Data licensing makes it clear to any person looking to reuse a dataset, who the data belongs to and how it can be used. Open licences (available to all) are often used to maximise data reuse, though different stipulations can be applied to these licences to restrict/enforce certain downstream activities, such as reuse of data for commercial purposes.

The Creative Commons licences are used commonly for life science data, of which there are 7: CC0, CC BY, CC BY-SA, CC BY-NC, CC BY-ND, CC BY-NC-SA, CC BY-NC-ND

CC0 is the most permissive licence where “no rights are reserved’. This means the data creator releases the data without restriction to the public domain.

The second CC BY licence is often written as CC BY 4.0 to reflect the most current (4th) version of the licence. Here, anyone can share and adapt data for any purpose, including commercial CC BY.

CC BY 4.0 is virtually identical to CC0, except by using CC BY, legal rights are retained to acknowledge the originator of the data.

Question

Look at the following site detailing the (Creative Commons licences)[https://creativecommons.org/share-your-work/cclicenses/]. What is the effect of adding the following letters to the CC BY licence; SA, NC and ND?

fair RNAseq licences. Open image in new tab

Figure 13: Screenshots from the Creative Commons licences definitions pages for SA, NC and ND.
Question

Look for the licensing options for arrayexpress at the bottom of the page under Licensing. Which licence is used?

fair RNAseq licence footer. Open image in new tab

Figure 14: The footer from the RNAseq data record showing the link to the Licensing conditions circled in red.

CC0 is used.

Question

How could I download the data or metadata from this page?

Machine-readable metadata can be downloaded from the “MAGE-TAB Files” link. There are 2 files: Investigation Design Format (experimental overview) and the Sample/Data Relationship Format (sample/data descriptions). Note this information can be accessed on the right hand banner as well.

Processed data can be downloaded also, in this case there is a link to a read-count matrix. Sequencing data from the NEA.

Access to raw FASTQ data is given via the reference to the ENA database on the right hand banner under “Linked Information”.

fair RNaseq metadata and rawdata. Open image in new tab

Figure 15: Links to metadata, raw data and processed data in the RNAseq record, circled in red.

(R1.2) (Meta)data are associated with detailed provenance.

The term “data provenance” relates to information about how and why a piece of data was produced. Often the individual producing the data and details of its production (protocols) are recorded as part of provenance.

Question

What provenance metadata can you find on Arrayexpress:E-MTAB-8316?

We are looking for information that could help us understand how data have been created. We have been given the name of the data creator noting that there is an option too for the author to submit an ORCID. Data creation and transformations are detailed under Protocols.

fair RNaseq submitter. Open image in new tab

Figure 16: Showing submitter and data provenance metadata in the RNAseq record, circled in red.