This is an automatic method that works for Cell phenotype classification. We integrate NIMBUS channel-wise scores with the biological cell-phenotype chart to classify the cell phenotype, rather than the traditional cluster method, which is unstable, requires human effort, and includes ambiguity. This project is inspired by NIMBUS and is a post-processing from Nimbus_repository.

Requirements

If you need the AI to formulate the cell phenotype lineage, you need to get an API KEY from Claude platform Claudeplate;

Otherwise, you can skip this step.

To enhance the LLM's phenotype knowledge, you can download the Cellmarker2 tables as they provide a huge number of marker functionalities across different tissues.

Installation

conda env create -f environment.yml

Folder structure

Set the file arc as below:

Agent

├── results

├── Cellmarker2

├── Chains

├── data

├── results

├── environment.yml

├── main.sh

├── scripts_.py

Running code

If you have run the code before, or you want to manually correct the phenotype family tree, you can go to the Chains folder, modify it, and use it directly; hence, you don't need the agent to search the database and generate the formatted JSON file. In this case, you need to input the DataFrame and the JSON file of the lineage.

main.sh 

If you are using the method at first time, and you want to just input your marker list, the species,s and the type of the image organ, you will need to get the phenotype lineage tree, and save the structure into a JSON file from the tree's bottom to top. In this case, you need to input the DataFrame and the biomarker list, species, and the tissue type; you need to have a CLAUDE API key.

main.sh --mode whole_pipline