SampleExplorer: using language models to discover relevant transcriptome data

Area of Science:

• Biomedical Research
• Bioinformatics
• Computational Biology

Background:

• Transcriptomics, particularly RNA-sequencing (RNA-seq), is a cornerstone of modern biomedical research.
• Large public repositories contain vast amounts of RNA-seq data with associated metadata, crucial for study understanding and replication.
• Existing metadata is underutilized for discovering relevant datasets, hindering efficient data reuse.

Purpose of the Study:

• To introduce SampleExplorer, a novel tool designed to facilitate the discovery of relevant transcriptomic datasets.
• To enable researchers to search for data using both text-based queries and gene set information.
• To improve the identification and accessibility of gene expression datasets within large public repositories.

Main Methods:

• SampleExplorer embeds sample metadata and utilizes a transformer-based language model for data retrieval.
• The tool was benchmarked using the ARCHS4 database to assess its effectiveness.
• Implementation details and algorithmic descriptions are available in supplementary materials.

Main Results:

• SampleExplorer effectively retrieves biologically relevant samples from large-scale transcriptomic data.
• The tool provides an efficient method for discovering relevant gene expression datasets.
• It enhances sample and dataset identification across diverse experimental contexts.

Conclusions:

• SampleExplorer offers a powerful solution for leveraging existing transcriptomic data.
• The tool supports replication and verification studies by improving data discoverability.
• It represents a significant advancement in utilizing RNA-seq metadata for research.