Annotation Boost (Optional)

Annotation Boost is an advanced validation tool that enhances annotation confidence through multiple iterations of analysis. It's particularly useful for:

• Validating low-confidence annotations
• Getting detailed insights into specific cell clusters
• Resolving ambiguous cell type assignments
• Generating comprehensive validation reports

Required Components

1. Input Data:

   • Full results CSV from CASSIA batch analysis
   • Original marker gene file (Seurat output or custom comlete marker file that contains all the statistics.)
   • Cluster context information
   • Specific cluster identifier

2. Model Configuration:

   • Recommended: anthropic/claude-3.5-sonnet via OpenRouter

Running Annotation Boost

# Setup parameters
validation_config <- list(
    model = "anthropic/claude-3.5-sonnet",
    provider = "openrouter"
)

# Define cluster information
cluster_info <- "Human PBMC"

#Specify the cluster you want to validate
target_cluster = "CD4+ T cell"

# Run validation
runCASSIA_annotationboost(

    # Required parameters
    full_result_path = "cell_type_analysis_results.csv",
    marker = marker_data,
    cluster_name = target_cluster,
    major_cluster_info = cluster_info,
    output_name = "Cluster1_report",
    num_iterations = 5, # Number of validation rounds

    # Model configuration
    model = validation_config$model,
    provider = validation_config$provider,
)
R

Parameter Details

• full_result_path: Path to original CASSIA results
• marker: Marker gene data (same as used in initial analysis)
• cluster_name: Target cluster name
• major_cluster_info: Dataset context
• num_iterations: Number of validation rounds (default: 5)

Troubleshooting

1. Low Confidence Results:

   • Review the quality of the clusters, in terms of doublet, mixed, or low quality clusters.
   • Review marker gene quality

2. Inconsistent Results:

   • Check marker gene consistency
   • Verify input data quality