An Ecosystem for Exploring the Universe of Single-Cell Data • celliverse

Clustering, marker discovery, cell-type annotation, and natural-language single-cell analysis in R

A robust and versatile toolkit for interpretable single-cell RNA-sequencing analysis.

Overview

CelliVerse is an R toolkit for single-cell RNA-sequencing (scRNA-seq) analysis, with integrated functionality for clustering, sub-clustering, marker discovery, cell-type annotation, cluster comparison, and visualization.

At its core, ClustoCell provides a data-driven framework for identifying major cellular populations and resolving biologically meaningful sub-populations while simultaneously generating ranked positive and negative marker profiles. CelliVerse also supports marker discovery for pre-defined clusters and custom cell subsets, curated marker-based annotation through CelliVerse MarkerDB, LLM-assisted annotation, and an integrated natural-language CelliVerse Agent.

CelliVerse is designed to be independent of library size and other sample- or cell-level confounding effects, supporting robust and interpretable analyses across diverse datasets.

At a glance

Capability	What CelliVerse provides
Clustering & sub-clustering	Data-driven identification of major clusters and biologically meaningful sub-clusters with `clustoCell()`
Marker discovery	Ranked positive and negative markers for clusters, sub-clusters, pre-defined groups, and selected cell subsets
Cell-type annotation	Curated MarkerDB-based annotation, direct LLM-assisted annotation, and portable LLM prompts
Cellular subset analysis	Marker discovery within user-defined subsets through the MarkoCell workflow
Visualization & comparison	Tools for inspecting clustering structure, marker profiles, annotations, and cluster relationships
Natural-language analysis	An integrated LLM-powered Agent for applying CelliVerse workflows through conversational instructions

Quick links

Resource	Link
✨ Feature Explorer	Explore CelliVerse capabilities
🧭 Adoption Hub	Interactive onboarding, function guidance, and troubleshooting
📦 CRAN	cran.r-project.org/package=celliverse
📖 Full vignette	CelliVerse documentation
🤖 CelliVerse Agent setup	Setup Guide
🤗 Interactive demo	Hugging Face Space
🧬 Reproducibility resources	CelliVerse-Project
🐞 Issues & feature requests	GitHub issue tracker

Installation

CRAN release

Install the current stable release from CRAN:

install.packages("celliverse")

Development version

Install the latest development version from GitHub:

# install.packages("remotes")
remotes::install_github(
  "asalavaty/celliverse",
  build_vignettes = TRUE
)

Then load the package:

library(celliverse)

Core workflows

ClustoCell

clustoCell() is the central clustering and marker-discovery workflow in CelliVerse. It supports identification of major clusters, sub-clustering of selected populations, and ranked positive and negative marker discovery within a unified analysis framework.

The resulting ClustoCell object can be used directly with downstream CelliVerse annotation, visualization, and prompt-generation tools.

Marker discovery with MarkoCell

The MarkoCell workflow complements ClustoCell when cluster identities or cell subsets are already defined. It enables marker discovery for pre-existing groups and custom-selected subsets without requiring cells to be re-clustered.

Cell-type annotation

CelliVerse provides several annotation routes so users can choose between curated references and LLM-assisted interpretation.

Route	Main interface	Best suited for
CelliVerse MarkerDB	`typoClust(mode = "markerDB")`	Reproducible annotation against the curated CelliVerse marker resource
Direct LLM annotation	`typoClust(mode = "ceLLMarkup")`	Annotation of ClustoCell results through a configured language model
Standalone LLM annotation	`ceLLMarkup()`	Annotation directly from marker panels or existing cluster labels
Portable annotation prompt	`typoPrompt()`	Generating a structured prompt for use with any preferred chatbot or LLM

The CelliVerse MarkerDB is also distributed with the package as the markerDB data object and contains harmonized positive and negative marker information used by the package’s curated annotation workflow.

🤖 CelliVerse Agent

CelliVerse includes an LLM-powered Agent that provides a browser-based natural-language interface to CelliVerse analyses. It is designed to make workflows such as ClustoCell clustering, marker discovery, cell-type annotation, and visualization more accessible to researchers who do not routinely write R code.

Start the Agent

install_celliverse_agent()  # one-time setup
run_celliverse_agent()      # launch the local browser app

The Agent can work with common single-cell data inputs, including:

.rds, .RData, and .rda;
.csv, .tsv, .txt, and .tab, optionally gzip-compressed;
Matrix Market .mtx files, optionally gzip-compressed, with their sidecar files;
.zip archives containing a 10x triplet; and
.h5 files when hdf5r is available.

Example requests include:

run clustoCell on the object

add labels to the Seurat object

generate a UMAP of the object and color cells by sub-clusters

give me the top 10 ranked markers of C2 and C4

annotate sub-clusters C1-Sub1 and C3-Sub2

The analysis objects and expression matrices remain in the local R session. When a cloud model is selected, the Agent sends only compact task-relevant information required for model reasoning rather than entire expression matrices. Fully local operation is also possible with providers such as Ollama and LM Studio.

Model capability varies. Lightweight or local models may perform best when requests are expressed as one clear task at a time, whereas more capable models generally handle multi-step requests more reliably.

For installation, provider configuration, security details, and advanced setup, see the CelliVerse vignette.

✨ Explore CelliVerse interactively

You can explore CelliVerse before installing the package through two complementary interactive resources.

✨ Feature Explorer

Explore the features, workflows, and capabilities of the CelliVerse R package through an interactive experience. This is a quick way to understand what CelliVerse can do and identify the workflows most relevant to your analysis.

🤗 ClustoCell interactive demo

Explore a PBMC3K ClustoCell analysis using precomputed outputs, including UMAP visualizations, ClustoCell clusters and sub-clusters, and ranked marker tables, without installing the package locally.

🧭 CelliVerse Adoption Hub

New to CelliVerse? The CelliVerse Adoption Hub provides a practical interactive guide for choosing the right workflow and function, following the minimal analysis route, understanding how the major functions connect, and resolving common first-use issues.

Documentation

A comprehensive introduction to CelliVerse and its workflows is available in the package vignette:

Read the CelliVerse vignette

You can also browse installed vignettes directly from R:

browseVignettes("celliverse")

Function-level documentation is available through standard R help:

?clustoCell
?typoClust
?ceLLMarkup
?typoPrompt

Reproducibility and study resources

The analysis scripts used to generate the figures associated with the CelliVerse/ClustoCell study, together with prepared CelliVerse MarkerDB resources, are available in the dedicated reproducibility repository:

github.com/asalavaty/CelliVerse-Project

Public bulk and single-cell RNA-seq datasets used in the study are also archived on Zenodo:

This separation keeps the R package lightweight while providing transparent access to the manuscript workflows and study resources.

How to cite CelliVerse

If you use CelliVerse, ClustoCell, CelliVerse MarkerDB, or related functionality in your work, please cite the associated CelliVerse/ClustoCell publication.

The full manuscript citation will be added upon publication.

You can also access the package citation information directly from R:

citation("celliverse")

Author

CelliVerse was developed by Adrian Salavaty.

Advisor

Ramyar Molania

Contributing and support

Bug reports, feature requests, documentation suggestions, and other contributions are welcome.

Please use the CelliVerse GitHub issue tracker to report problems or suggest enhancements.

When reporting a bug, including a minimal reproducible example together with your R and package versions will make it easier to diagnose the issue.