CellAtlasSearch: a scalable search engine for single cells.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput single-cell transcriptomics generates vast amounts of gene expression data.
• Existing efforts focus on homogenizing and storing this data, but making it searchable is key to unlocking its value.
• Large-scale gene expression datasets require efficient search architectures.

Purpose of the Study:

• To develop a novel, scalable, and searchable architecture for high-dimensional gene expression data.
• To enable efficient querying and characterization of unannotated single cells.
• To facilitate noise-free, low-dimensional representation of single-cell expression profiles.

Main Methods:

• Development of CellAtlasSearch, a massively parallel and lightweight search architecture.
• Implementation of a Graphical Processing Unit (GPU)-friendly Locality Sensitive Hashing (LSH) for accelerated data processing and querying.
• Inclusion of over 300,000 reference expression profiles (bulk and single-cell data).

Main Results:

• CellAtlasSearch offers unmatched speedup in data processing and query through GPU-friendly LSH.
• The system successfully queries individual single-cell transcriptomes and retrieves matching samples with meta-information.
• Provides a resource for characterizing unannotated cells and generating low-dimensional representations.

Conclusions:

• CellAtlasSearch provides a scalable and efficient solution for searching large-scale gene expression datasets.
• The architecture facilitates the analysis and characterization of single-cell data for researchers and clinicians.
• Enables deeper insights into biological systems through accessible and searchable transcriptomic data.