biorag 0.1.1

BioRAG: A tool for textual and gene set search against ARCHS4 data

BioRAG

This is the repository for BioRAG[^1]. BioRAG can identify relevant studies within the ARCHS4 database, using a text-based query or a gene set.

Table of Contents

• Installation
• Usage
• License
• References

Installation

Install BioRAG via PyPI with the following command:

pip install biorag

Additional files

These additional files will need to be downloaded. Below is a table containing a description of these files, and the links.

File Name	Description	Size	checksum
semantic_db.h5ad	semantic vector store	58Mb	bd23f8835032dcd709f8ca05915038b3
transcriptomic_db.h5ad	transcriptomic vector store	38GB	d26e564653424ed6e189aea7d0be5d4d
human_gene_v2.2.h5	ARCHS4[^2] hdf5 database	37GB	f546fdecc0ba20bf2a5571628f132ca5

Both the semantic vector stores and the transcriptomic vector stores are AnnData[^3] objects, containing both the embedding matrices and the indices which link each embedding vector to an experiment in the ARCHS4 database. Additionally, the transcriptomic vector store also contains derived count data of "representative transcriptomes" in the ARCHS4 database.

Usage

To load BioRAG, follow these steps:

1. Download the additional files described above.

2. Instantiate a query_db object to perform BioRAG search in following way:

   from biorag.biorag import Query_DB
   
   new_query_db = Query_DB(SEMANTIC_VECTOR_STORE_PATH, TRANSCRIPTOME_VECTOR_STORE_PATH, ARCHS4_HDF5_DATABASE_PATH)
   

3. Run BioRAG with a gene set query and/or textual query. The gene set is a python list, and the textual query is a python string.

   text_query = "This text query usually describes the experiment"
   
   geneset_query = ["IFNG", "IRF1", "IFR2"]
   
   result = new_query_db.search(geneset = geneset_query, text_query = text_query)
   

4. The output is a Results object, which is composed of three pandas dataframes, which can be accessed via dot notation.

• The "seed_studies" variable holds a dataframe containing study metadata from the search step.
• The "expansion_studies" holds a dataframe of studies from the expansion step,
• The "samples" variable holds a dataframe of the relevant samples, with metadata derived from the ARCHS4 database.

    result = new_query_db.search(geneset = geneset_query, text_query = text_query)

    # save the results as csv files
    result.seed_studies.to_csv("relevant_seed_studies.csv")
    result.expansion_studies.to_csv("relevant_expansion_studies.csv")
    result.samples.to_csv("relevant_samples.csv")

Modifying searches using different inputs

All BioRAG searches should contain at least a text query or a gene set query. Users can choose one of several search strategies, with examples illustrated below:

• Strategy 1: Text only as input

  result = new_query_db.search(text_query = text_query, geneset = None)
  

• Strategy 2: Text and gene set as input

  result = new_query_db.search(text_query = text_query, geneset = geneset_query)
  

• Strategy 3: Gene set only as input

  result = new_query_db.search(text_query = None, geneset = geneset_query)
  

Modifying searches using different expansion strategies

By default, BioRAG will perform semantic search, followed by transcriptomic expansion. However, this search strategy can be modified. For instance, one can use semantic search, followed by perfoming the expansion step using semantic similiarity. In this case, the transcriptome vector store is not queried.

The "seed" and "expand" parameters accepts either "transcriptome" or "semantic" as options.

If transcriptome search is performed as the initial step, the count data from "representative transcriptomes" in the transcriptome vector store is used to search for enriched samples, using ssGSEA (single sample gene set enrichment analysis) to rank the most relevant samples. In semantic search, BioRAG retrieves the most semantically similar studies to the text query, using cosine distance as the metric.

1. Example 1 - perform semantic search, followed by transcriptome expansion.

   result = new_query_db.search(text_query = text_query, geneset = geneset_query, search = "semantic", expand = "transcriptome")
   

2. Example 2 - perform transcriptome search, followed by semantic expansion.

   result = new_query_db.search(text_query = text_query, geneset = geneset_query, search = "transcriptome", expand = "semantic")
   

3. Example 3 - perform semantic search, followed by semantic expansion, using only a text query.

   result = new_query_db.search(text_query = text_query, geneset = None, search = "semantic", expand = "semantic")
   

The types of searches which can be performed depend on input. For instance,if only a gene set is supplied, the search step defaults to "transcriptome", with the user able to select between "transcriptome" and "semantic" for the expansion step.

Optional single sample gene set enrichment analysis (ssGSEA)

To further refine the set of samples and studies returned by BioRAG search, ssGSEA can be peformed on all samples returned by the query. Use the "perform_enrichment" parameter to specifiy if ssGSEA should be performed on all samples. If so, the returned dataframe will contain enrichment scores, pvalues and FDRs. The ssGSEA results will be stored as a dataframe in the "samples" attribute in the Results object.

    result = new_query_db.search(text_query = text_query, geneset = None, search = "semantic", expand = "semantic", perform_enrichment = True)

    # save enrichment results from the sample dataframe
    results.samples.to_csv("samples_with_ssgsea_results.csv")

License

Creative Commons Attribution 4.0 International. The Creative Commons Attribution license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited.

References

[^1]: Chin WL, & Lassmann, T. (2024). Language models improve the discovery of public RNA-seq data. bioRxiv. [^2]: Lachmann A, Torre D, Keenan AB, Jagodnik KM, Lee HJ, Wang L, Silverstein MC, Ma'ayan A. Massive mining of publicly available RNA-seq data from human and mouse. Nat Commun. 2018 Apr 10;9(1):1366. doi: 10.1038/s41467-018-03751-6. PMID: 29636450; PMCID: PMC5893633. [^3]: [AnnData] (https://anndata.readthedocs.io/en/latest/)