henhaobio
Sign in

bioRxivImmunologyPreprint

CellKb Immune: a manually curated database of mammalian hematopoietic marker gene sets for rapid cell type identification

Ajay Patil, Ashwini PatiliD

PDFDOIData
In-depth full-text digest available · Jump to digest ↓

Abstract

Using single-cell RNA sequencing, it is now possible to get the detailed gene expression pattern in each cell for millions of cells in a single experiment under varying conditions. Annotating cells by assigning a type or state to them is important for correct interpretation of results. In spite of current advances in technology, assigning cell types in single-cell datasets remains a bottleneck due to the lack of a comprehensive reference database and an accurate search method in a single tool. CellKb Immune is a knowledgebase of manually collected, curated and annotated marker gene sets representing cell types in the immune response in mouse. It allows users to find matching cell types in literature given a list of genes. We evaluated the contents and search methodology of CellKb Immune using a leave-one-out approach. We further used CellKb Immune to annotate previously defined marker gene sets from a bulk RNA-seq dataset to confirm its accuracy and coverage. CellKb Immune provides an easy-to-use online tool with a reliable method to find matching cell types and annotate cells in single-cell experiments. It is available at https://www.cellkb.com/immune.

In-depth summary

Claude Opus 5.5 · xhigh effort · 2 Nov 2026 · from the full text of version 5 (13 Feb 2022)

CellKb Immune curates 2,395 mouse hematopoietic marker gene sets from 226 publications and uses a rank-based search algorithm to annotate single-cell RNA-seq clusters via a web interface.

Annotating cell types in single-cell RNA-seq experiments typically requires manual literature searches or programmatic integration of limited reference datasets, leaving most published marker signatures unused. Unlike CellMarker (single-gene queries) and PanglaoDB (unranked queries against aggregated signatures), CellKb Immune preserves author-defined ranked marker gene sets with standardized ontologies and searches them directly using ranked gene lists.

Key results

  1. ShownCellKb Immune compiles 2,395 manually curated mouse marker gene sets covering 230 hematopoietic cell types and 104 tissues or organs from 226 publications, alongside protein-protein interaction networks.

    Selected from approximately 10,000 studies published between 2013 and 2021, the database stores up to 1,500 genes per set (corrected p-value <= 0.1, log fold change > 0) mapped to Ensembl identifiers, Cell Ontology (CL:0000988 and descendants), Uberon Ontology, Disease Ontology, and 40,122 HitPredict protein-protein interactions. · 2.1 Data collection and curation; 2.2 Data harmonization and annotation; 3.1 Database contents

  2. ShownLeave-one-out cross-validation across ontologies demonstrates that CellKb Immune's rank-based search method accurately retrieves matching or closely related hematopoietic cell types within the database.

    Querying each of 2,223 ontology-annotated mouse marker gene sets against all remaining sets in CellKb Immune matched the exact same Cell Ontology term or its parent as the top hit for 1,568 sets (70.5%) and a sibling ontology term for 427 sets (19.2%), totaling 89.7% across same, parent, or sibling terms. · 3.3 Evaluation; 10 Supplementary Material; Tables

  3. ShownCellKb Immune accurately annotates external bulk RNA-seq immune cell signatures from the ImmGen consortium.

    Using the top 1,500 significantly upregulated genes for 98 marker gene sets representing 40 distinct cell types from Immgen bulk RNA-seq data, the first hit in CellKb Immune matched the exact ontology or its parent in 61 instances (62.2%) and a sibling ontology in 18 instances (18.4%), totaling 79.5%. · 3.3 Evaluation; 10 Supplementary Material; Tables

Methods, data and reproducibility

Author-defined mouse marker gene sets from published scRNA-seq, bulk RNA-seq, and microarray studies were manually curated from PubMed, filtered (corrected p-value <= 0.1, log fold change > 0, up to 1,500 genes), and mapped to Ensembl IDs and Cell, Uberon, and Disease ontologies. A web tool scores ranked query gene lists against all database sets using shared gene counts, ranks, rank differences, and set sizes.

  • Curated mouse hematopoietic marker gene sets from scRNA-seq, bulk RNA-seq, and microarray literature (including MSig-db signatures) published between 2013 and 2021 · mouse · n = 2,395 marker gene sets (2,223 ontology-annotated) from 226 publications
  • Protein-protein interactions with reliability scores from the HitPredict database integrated into cell-type-specific networks · mouse · n = 40,122 protein-protein interactions
  • External validation bulk RNA-seq dataset of mouse immune cell types from Immgen · mouse · n = 98 marker gene sets representing 40 distinct cell types
Methods & tools
CellKb Immune, Ensembl, Cell Ontology, Uberon Ontology, Disease Ontology, HitPredict, MSig-db, Immgen, CellMarker, PanglaoDB
Reproducibility gaps
Source code, explicit mathematical formulas for the rank-based match score and reliability score, and bulk database download files are not provided; access is via the web interface.
Limitations (2)
  • Despite 'mammalian' in the title, the described CellKb Immune database and evaluations are restricted exclusively to mouse hematopoietic cell types. · noted in this summary
  • Quantitative benchmarking of cell type assignment accuracy is performed only internally (leave-one-out) and on one ImmGen bulk RNA-seq dataset, without head-to-head accuracy comparisons against CellMarker, PanglaoDB, or automated scRNA-seq annotation tools. · noted in this summary

Read this if you annotate mouse immune or hematopoietic cell clusters from single-cell or bulk RNA-seq data and want a web tool to search ranked gene lists against literature-curated marker sets.

Citations
25

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLiu '21Aybey '23Cheng '24Wang '20Mao '21Xu '21
65% semantic matchBioinformatics
Xuan Liu, Sara J.C. Gosline, Lance T. Pflieger+5 more·Briefings in Bioinformatics·Feb 3, 2021· 16 citations