Polly vs Recount3 - Comparing Findability for RNA-seq Data

Polly vs Recount3 - Comparing Findability for RNA-seq Data

March 7, 2023

RNA sequencing is a rapidly emerging method for investigating the transcriptome. Over the past few decades, it has significantly progressed, becoming a paramount approach in transcriptome profiling. RNA-seq data is being utilized in multiple aspects of research and disease treatments. However, findability, usability, quality, and reliability have always been problematic for researchers and data scientists.  

Though this is a very niche space, multiple platforms are being developed to facilitate the availability of  RNA-seq data and do so in varying degrees of efficiency. Here we lay out a comparison between two such platforms operating in a similar space.

In this blog, we compare and discuss the difference between Elucidata’s ML-Ops platform Polly, and an online resource Recount3, as sources for uniformly processed and annotated RNAseq data.

What is Polly?

Polly is a data-centric MLOps platform that hosts FAIR (Findable Accessible Interoperable and Reusable) multi-omics data from public and proprietary sources. Specific ETL pipelines called Connectors facilitate seamless data ingestion and harmonization. Polly’s curation infrastructure is built on a specialized BERT model, PollyBERT, that helps in metadata annotation.

What is Recount3?

Recount3 is an online resource that consists of uniformly processed RNA-seq data. It consists of RNA-seq gene, exon, and exon-exon junction counts as well as coverage bigWig files for 8,679 and 10,088 different studies for humans and mouse respectively. It is the third generation of the ReCount project and part of recount.bio.The raw sequencing data is processed with the Monorail system which generates the coverage bigWig files and the recount-unified text files. Furthermore, snapcount enables query-based access to the recount3 and recount2 data.

Polly v/s Recount3: A Comparison

Let us dive deeper into understanding how these platforms work with the help of a few examples.

1. Querying Efficiency

Querying at GUI level for transcriptomics datasets Neurodegenerative diseases in humans.

# of datasets found:
Time taken
  • Availability of elaborate set of filters that improves dataset findability.

  • The search automatically expands to include diseases related to the term neurodegenerative disease such as AD, Huntingtons, Parkinsons etc

  • All datasets have data files associated with them ready to download

  • Gives results from all 3 of its sources in single search

Querying programmatically for Alzheimer's disease datasets with normal and patient samples.

# of datasets found:
10 42
Time taken
<5 minutes
>45 minutes

  • Recount does not offer programmatic querying capabilities across datasets.

  • To identify relevent datasets for the query used above, the user needs to use the study explorer and read through the description of the datasets to get datasets with Normal and disease samples.

2. How Easy Is It to Find Relevant Data?

  • Polly GUI and Polly-python both support keyword searches through metadata fields following a standard ontology across all the datasets. It also allows free text searches.
  • Recount3 does not allow searching datasets programmatically using keyword searches. There is no metadata curation, hence the metadata ontologies differ across datasets, making it difficult to find similar datasets using a single keyword.

3. How Easy Is It to Access the Data?

  • Polly platform and Polly-python are proprietary software requiring a license for their usage and hence also the data hosted on the platform.
  • On Recount3, the data and R package is open source, hence no authorization is required to access the data.

4. How Easy Is It to Integrate the Data with Other Data and Interoperate with Applications or Workflows for Analysis, Storage, and Processing?

  • On Polly, all the datasets are processed through the same Kalisto pipeline and the dataset metadata follow a standard ontology allowing easy comparison of datasets. They can be easily used with downstream analysis packages.
  • On Recount3, all datasets are processed through the same Monorail pipeline and can be easily used with downstream analysis packages. However, the lack of standard ontology is a big problem since you might miss out on data just because of the difference in vocabulary.

Data FAIRness Comparison    

Ratings out of 5

⭐️⭐️⭐️⭐️⭐️ ⭐️⭐️
Interoperability ⭐️⭐️⭐️⭐️⭐️


Comparing Data Availability and Usability  

  • Comparing the volume, variety, and sources of the data
  • Comparing dataset processing

  • Bulk RNASeq, sc RNASeq
  • Microarray, Bulk RNASeq, sc RNASeq (only smartseq platform)
Data sources
  • GEO, SCP, Human cell Atlas, Single Cell Expression Atlas, Tabula Sapiens, HTAN, Zenodo, Covid-19 cell Atlas
Data Volume
  • 735,914 RNA-seq samples across all organisms. 
  • 41,638 datasets from human and mouse studies.
  • New datasets are added regularly.
  • 750,000 human and mouse RNA-seq samples. 
  • 18,767 datasets human and mouse studies.
  • New datasets are not added regularly.
  • Human, mouse, rat, primates and other organisms

  • Human, Mouse only
Data processing
  • All bulk RNA seq datasets are processed through Kalisto based pipeline, if the raw data is available for the dataset,following the best published RNASeq practices 
  • Datasets without available raw data are ingested as is from the source with metadata standardization.
  • All datasets are processed thorugh their distributed processing system Monorail 
  • Monorail uses STAR and related tools to summarize expression at the gene and exons levels (annotation-dependent), to detect and report exon-exon splice junctions, and to summarize coverage along the genome as a bigWig file .

While both platforms are great sources for finding processed RNAseq data, it would be helpful to take a closer look to identify how they would serve particular users. It is very important for researchers and scientists to keep up with all the emerging data without having to spend a lot of time finding the relevant ones. It is thus preferable to have metadata backed with standard ontologies enabling superior search and findability. We hope this blog can help users make an informed choice between these platforms.

If you are spending time scouring datasets to just find out relevant ones for downstream analysis, now is the time to reach out. Connect with us to learn more about how to accelerate your research.

Request Demo