CASSIA

CASSIA (Cell type Annotation using Specialized System with Integrated AI) is a Python package for automated cell type annotation in single-cell RNA sequencing data using large language models.

Features

• Automated cell type annotation using multiple LLM providers (OpenAI, Anthropic, OpenRouter)
• Support for batch processing of multiple clusters
• Variance analysis for annotation reliability
• Detailed HTML report generation
• Score-based annotation quality assessment
• Support for marker gene validation
• Interactive analysis with step-by-step reasoning

Installation

#Prepare dependency
pip install pandas openai requests numpy anthropic

#Install CASSIA
pip install CASSIA
pip install CASSIA_rag  # optional for the RAG agent

Quick Start

import CASSIA

# Set your API keys
CASSIA.set_api_key("your-openai-key", provider="openai")
CASSIA.set_api_key("your-anthropic-key", provider="anthropic")
CASSIA.set_api_key("your-openrouter-key", provider="openrouter")

# Run single cluster analysis
result = CASSIA.run_cell_type_analysis_wrapper(
    model="gpt-4",
    temperature=0,
    marker_list=["CD3D", "CD4", "IL7R"],
    tissue="PBMC",
    species="human",
    provider="openai"
)

# Run batch analysis
results = CASSIA.run_cell_type_analysis_batchrun(
    marker="path/to/marker_file.csv",
    output_name="results.json",
    model="gpt-4",
    tissue="PBMC",
    species="human"
)

Input Formats

CASSIA accepts marker genes in several formats:

• CSV files with cluster and marker columns
• Pandas DataFrames
• Direct lists of marker genes
• Seurat or Scanpy differential expression results

Key Functions

Basic Analysis

• run_cell_type_analysis_wrapper(): Single cluster analysis with support for multiple LLM providers
• run_cell_type_analysis_batchrun(): Batch analysis of multiple clusters
• run_analysis_n_times(): Multiple runs for consensus analysis

Advanced Features

• process_cell_type_variance_analysis_batch(): Analyze annotation variance
• score_annotation_batch(): Score annotation quality
• generate_cell_type_analysis_report_wrapper(): Generate detailed HTML reports

Utility Functions

• set_api_key(): Set API keys for different providers
• get_top_markers(): Extract top markers from differential expression results
• split_markers(): Process marker lists into standardized format

Detailed Usage Examples

Single Cluster Analysis

# Basic single cluster analysis
result, conversation = CASSIA.run_cell_type_analysis_wrapper(
    model="gpt-4",
    temperature=0,
    marker_list=["CD3D", "CD4", "IL7R", "FOXP3"],
    tissue="PBMC",
    species="human",
    provider="openai"
)

# Print the main cell type identified
print(result["main_cell_type"])

Batch Analysis with Multiple Iterations

# Run batch analysis multiple times for consensus
CASSIA.run_batch_analysis_n_times(
    n=3,  # Number of iterations
    marker="markers.csv",
    output_name="batch_results",
    model="gpt-4",
    tissue="lung",
    species="human",
    max_workers=10
)

Generate Analysis Report

CASSIA.generate_cell_type_analysis_report_wrapper(
    full_result_path="results.csv",
    marker="markers.csv",
    cluster_name="cluster_1",
    major_cluster_info="Human PBMC",
    output_name="analysis_report.html",
    model="gpt-4",
    provider="openai"
)

Contributing

We welcome contributions! Please feel free to submit pull requests or open issues on our GitHub repository.

License

This project is licensed under the MIT License.

Support

For support, please open an issue on our GitHub repository.