- Python 47%
- R 28.9%
- Shell 24.1%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| data | ||
| prompt | ||
| python-liana | ||
| python-llm | ||
| Rscript | ||
| screenshots | ||
| tutorial | ||
| .DS_Store | ||
| .gitignore | ||
| iS2C2.sh | ||
| LIANAPlus-llm.sh | ||
| NicheNet-llm.sh | ||
| parameters.md | ||
| pbmc_control_example_clean_2_4_26.h5ad | ||
| README.md | ||
| requirements.txt | ||
iS2C2: A Co-Intelligent Platform for Mechanistic Discovery of Disease Cellular Crosstalk
Table of Contents
- iS2C2: A Co-Intelligent Platform for Mechanistic Discovery of Disease Cellular Crosstalk
Introduction
This repository provides a fully automated pipeline for hypothesis generation from cell-cell communication (CCC) data using large language models (LLMs), built for scalability, transparency, and ease of use. The current version supports single-cell and spatial transcriptomics data, including 10x datasets for human and mouse samples.
Unlike traditional approaches that require manual interpretation of omics data outputs, our pipeline combines computational algorithms with LLMs to automate the generation of interpretable biological hypotheses.
Key Features
-
✅ Fully Automated Workflows with API Integration
The pipeline is optimized for completely automated execution via local OpenModels (e.g., Ollama) and API Call Functions (e.g., Gemini, OpenRouter), streamlining the entire process from input data to hypothesis generation.
Note: Chatbox interfaces for ChatGPT or KIMI can be used but require some manual steps for interaction. -
✅ Example Datasets and Transparent Tutorials
Includes curated example datasets and step-by-step tutorials with clear outputs at every stage, enabling easy testing, customization, and reproducibility. -
✅ Multi-Source CCC Compatibility
Seamlessly integrates outputs from multiple CCC pipelines such as S2C2, LIANA+, and NicheNet, enabling flexible use of various upstream analyses before LLM-powered hypothesis generation. -
✅ Fully compatible with both single-cell and spatial transcriptomics datasets
Supports seamless analysis and hypothesis generation for both data types without additional configuration.
Overview
This platform bridges the gap between structured computational data analysis and generative AI reasoning. By automating both data processing and biological interpretation steps, it enables:
- Scalable and reproducible hypothesis generation
- Domain-specific interpretability
- Supports both single-cell and spatial transcriptomics data
This approach was optimized for accuracy, repeatability, and interpretability. We validated our findings using two use cases: one focused on Alzheimer's disease and the other on cancer datasets.
Disclaimer: The results generated by this algorithm may include speculative or AI-generated associations. Users are advised to evaluate all outputs through independent literature review and critical analysis.
Update
[2025.8.1] The current iS2C2 supports S2C2 with LLM interpretation, LIANA+ with LLM interpretation, and NicheNet with LLM interpretation. The current version supports single-cell and spatial transcriptomics data for both human and mouse samples, and will generate 3 hypotheses for each cell-cell communication prediction.
System Requirements
| Component | Minimum | Recommended |
|---|---|---|
| RAM | 8GB | 16GB+ |
| Storage | 10GB | 50GB+ |
| CPU | 4 cores | 8+ cores |
| GPU | Optional | NVIDIA GPU for faster LLM inference |
Platform Compatibility: This platform is optimized for macOS and Linux systems. Windows users are recommended to use the S2C2 GUI interface and chatbox functionality for optimal performance.
Installation
git clone https://github.com/methodistsmab/iS2C2.git
cd iS2C2
LLM tool package install
Use Python 3.9
conda create -n is2c2 python=3.9
conda activate is2c2
Install Python packages
pip install -r requirements.txt
Ollama download & Install
- Visit the Ollama official website
- Download the installer for your operating system
- Follow the installation instructions included with the downloaded installer
Tutorial
Running cell-cell interaction using iS2C2
Running cell-cell interaction using LIANA+ and LLMs
LIANA+ supports two modes of operation:
- Mode 1 (Recommended): Run LIANA analysis directly from raw
.h5addata by providing--inputand--method, then automatically generate LLM hypotheses. - Mode 2: Provide a pre-computed LIANA result CSV via
--liana-result.
Quick Start (Mode 1 — Ollama):
./LIANAPlus-llm.sh \
--input ./pbmc_control_example_clean_2_4_26.h5ad \
--method logfc \
--cell-type "astrocyte-excitatory neuron" \
--disease-context "Alzheimer's disease" \
--llm-provider "ollama" \
--model "qwen3:30b"
- LIANA+ with local OpenModel (Ollama as the example)
- LIANA+ with API Call Function
- LIANA+ with Chatbox
Running cell-cell interaction using NicheNet and LLMs
- NicheNet with local OpenModel (Ollama as the example)
- NicheNet with API Call Function
- NicheNet with Chatbox
Expected Output
The final report is generated in the file:
llm_report_*.txt
This file contains a structured summary of the hypotheses generated by the pipeline.
✅ Example Report Output
Acknowledgements
We are deeply grateful for the open source software that powers our research and this project.
Citation
If you use iS2C2-llm cointelligence in your research, please cite:
@article{Sheng2025iS2C2,
title={{iS2C2: A Co-Intelligent Platform for Mechanistic Discovery of Disease Cellular Crosstalk}},
author={Jianting Sheng, Juyoung Ahn, Li Yang, Zhihao Wan, et al.},
journal={{[Journal Name - To be filled upon publication]}},
year={2025}
}
