EM-H01-human-single-neuron: LLM-Based Curation for Single-Cell Morphologies from the electron microscopic dataset H01
This repository contains the official codebase for the automated curation and evaluation of the human cortical neuron electron microscopy (EM) dataset (H01). It introduces an LLM-based system designed to advance single-cell neuroinformatics by utilizing structural priors to automatically generate, execute, and evaluate morphological curation scripts.
This work is associated with the publication:
Multimodal Data Fusion Reveals Morpho-Genetic Variations in Human Cortical Neurons Associated with Tumor Infiltration. Yufeng Liu et al., 2025. DOI: 10.64898/2025.12.26.696632
The codebase is primarily organized into two functional pipelines:
llmauto1/(Prior-Enhanced Script Factory): The core LLM-based generation system. This module utilizes established topological priors to programmatically generate custom curation and pruning scripts tailored to specific neuronal morphological anomalies. It was a compelling system in the ChatGPT-4/4o era, but feels somewhat dated in the age of ChatGPT-5.5 and OpenClaw.- Execution & Evaluation Scripts: The remainder of the repository contains the necessary infrastructure to execute the auto-generated scripts from
llmauto1against the raw.swcfiles. It also includes quantitative evaluation pipelines to assess the accuracy, topological validity, and overall quality of the refined reconstructions.
The corresponding curated morphological dataset, containing the reconstructed human cortical neurons derived from the H01 volume, is hosted publicly on Zenodo.
- Dataset Download: 10.5281/zenodo.19853169
(Note: Please refer to the inline comments within individual scripts for specific argument requirements and environment setups).
-
Clone the repository:
git clone [https://github.com/SEU-ALLEN-codebase/EM-H01-human-single-neuron.git](https://github.com/SEU-ALLEN-codebase/EM-H01-human-single-neuron.git) cd EM-H01-human-single-neuron -
Generate Curation Scripts: Navigate to the
llmauto1directory to configure and run the LLM script factory based on your specific structural priors. -
Execute and Evaluate: Run the generated scripts on your local directory of
.swcfiles, then utilize the root evaluation scripts to validate the pruning and branch-merging steps.
This project is mainly developed by Xiaoqin Gu before her graduation from Southeast University in 2025.
If you utilize this codebase or the associated H01 curated dataset in your research, please cite our paper:
@article{liu2025multimodal,
title={Multimodal Data Fusion Reveals Morpho-Genetic Variations in Human Cortical Neurons Associated with Tumor Infiltration},
author={Liu, Yufeng and others},
year={2025},
doi={10.64898/2025.12.26.696632}
}