### Install Dependencies

Source: https://github.com/yrsong001/gptanno/blob/master/PDF2markers/README.md

Install the required Python packages for the project.

```bash
pip install -r requirements.txt
```

```bash
pip install pandas numpy pypdf tqdm openai requests
```

--------------------------------

### Install GPTAnno

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Install the package from GitHub using devtools.

```r
# Install devtools if you haven't already
install.packages("devtools")

# Install GPTAnno
devtools::install_github("yrsong001/GPTAnno")
```

--------------------------------

### Initialize Ollama Server

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Commands to pull a model and start the local Ollama server.

```bash
# In a terminal:
ollama pull llama3    # or mistral, neural-chat, etc.
ollama serve          # default: http://localhost:11434
```

--------------------------------

### Ollama Setup Commands

Source: https://github.com/yrsong001/gptanno/blob/master/man/call_llm.Rd

Commands to prepare a local Ollama environment before calling the function.

```bash
ollama pull llama2       # or another model
ollama serve             # start the server (runs on localhost:11434)
```

--------------------------------

### Configure Alternative LLM Providers

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.html

Examples for configuring Anthropic or Gemini as the LLM backend.

```R
llm_config <- list(provider = "anthropic", model = "claude-sonnet-4-20250514")
llm_config <- list(provider = "gemini", model = "gemini-2.0-flash")
```

--------------------------------

### Build CL Term Map Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/build_cl_term_map.Rd

Demonstrates how to initialize an ontology object and generate the term map.

```R
# cl <- ontologyIndex::get_ontology(
#   "http://purl.obolibrary.org/obo/cl.obo",
#   extract_tags = "everything"
# )
# cl_term_map <- build_cl_term_map(cl)
```

--------------------------------

### Score Annotation Distance Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/score_annotation_distance_ontology.Rd

Demonstrates the setup of the ontology graph and the subsequent calculation of annotation distances.

```R
# Build ontology and graph (one-time setup)
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo",
 extract_tags = "everything")
graph <- build_ontology_graph(cl)
cl_term_map <- build_cl_term_map(cl) # if need the up-to-date mapping
# Score annotation distance
dist_res <- score_annotation_distance_ontology(
  seurat_obj, manual_col = "celltype_manual", predicted_col = "celltype_parent",
  cl_term_map = cl_term_map, # leave it blank will use the built-in mapping
  graph = ontology_graph
)
print(dist_res$mean_distance)
```

--------------------------------

### Configure Ontology Annotation Strategy

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_ontology_strategy.Rd

Examples demonstrating how to initialize the ontology object and configure annotation strategies using different restriction combinations.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo")

# Use only ontology descendants
config <- prepare_ontology_strategy(cl, "T cell")

# Combine descendants with user list
config <- prepare_ontology_strategy(
  cl, "T cell",
  user_restrict_to = c("regulatory T cell", "gamma-delta T cell"),
  combine_restrictions = TRUE
)

# Use ONLY user list (ignore descendants)
config <- prepare_ontology_strategy(
  cl, "T cell",
  user_restrict_to = c("regulatory T cell", "effector T cell"),
  combine_restrictions = FALSE
)
```

--------------------------------

### Run Multi-Resolution Clustering Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/run_multi_resolution_clustering.Rd

Demonstrates the basic usage of the function with a Seurat object and a vector of resolution parameters.

```R
# result <- run_multi_resolution_clustering(seurat_obj, c(0.1, 0.2, 0.3))
```

--------------------------------

### Example of user_restrict_to parameter usage

Source: https://github.com/yrsong001/gptanno/blob/master/man/run_subcluster_annotation_workflow.Rd

Demonstrates how to define per-celltype restrictions using a named list.

```R
list("T cell" = c("regulatory T cell", "effector T cell"),
                      "B cell" = c("memory B cell", "naive B cell"))
```

--------------------------------

### Configure Alternative LLM Providers

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Example of using an Anthropic model with specific decoding parameters.

```r
results <- gptanno(
  seurat_obj = your_seurat_object,
  resolutions = c(0.1, 0.3, 0.5),
  cl = cl,
  graph = graph,
  tissue_name = "mouse heart",
  llm_config = list(
    provider = "anthropic",
    model = "claude-sonnet-4-20250514",
    params = ellmer::params(temperature = 0.2, top_p = 0.9)
  ),
  marker_dir = marker_dir,
  n_runs = 2
)
```

--------------------------------

### Calculate Mean Ontology Distance Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/calculate_mean_ontology_distance.Rd

Basic usage example for calculating mean shortest path distance between CL IDs.

```R
# res <- calculate_mean_ontology_distance(clid_df, graph)
```

--------------------------------

### Clean and Match Cell Type Annotations

Source: https://github.com/yrsong001/gptanno/blob/master/man/clean_and_match_annotation.Rd

Examples demonstrating the standardization of various cell type strings.

```R
clean_and_match_annotation("t cell")
clean_and_match_annotation("Helper T cells")
clean_and_match_annotation("unknown cell")
```

--------------------------------

### Extracting Markers from PDFs

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Commands to install dependencies, configure the OpenAI API key, and execute the extraction and filtering pipeline.

```bash
cd pdf2markers

# Install Python dependencies
pip install -r requirements.txt

# Set OpenAI API key
export OPENAI_API_KEY="your-api-key"

# Extract markers from a paper
python paper_extraction_cellNgenes.py --pdf "papers/example.pdf" --out "outputs"

# Filter out ontology-annotated cell types
python filterout_cell_ontology.py --input-dir ./outputs
```

--------------------------------

### Score Annotation Agreement with Detailed Match Types

Source: https://github.com/yrsong001/gptanno/blob/master/man/score_annotation_agreement_ontology_detailed.Rd

Example usage of the function, including the necessary one-time setup of the ontology graph and ancestor map.

```R
# Setup ontology (one-time)
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo",
                                  extract_tags = "everything")
graph <- build_ontology_graph(cl)
cl_term_map <- build_cl_term_map(cl)
ancestor_type_map <- build_ancestor_type_map(cl)

# Score annotations with detailed match types
results <- score_annotation_agreement_ontology_detailed(
  seurat_obj,
  manual_col = "celltype_manual",
  predicted_col = "celltype_predicted",
  cl_term_map = cl_term_map,
  cl_ontology = cl,
  graph = graph,
  ancestor_type_map = ancestor_type_map,
  output_csv = "detailed_scores.csv"
)

# View summary
print(results$summary)
```

--------------------------------

### Preprocess Seurat Object Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/preprocess_seurat_object.Rd

Demonstrates the basic usage of the function to preprocess a Seurat object and optionally save the result to an RDS file.

```R
# seurat_obj <- preprocess_seurat_object(seurat_obj, save_path = "preprocessed_seurat.rds")
```

--------------------------------

### Run Subclustering and Marker Identification

Source: https://github.com/yrsong001/gptanno/blob/master/man/subcluster_and_find_markers.Rd

Example usage of the subcluster_and_find_markers function, including loading the required ontology and executing the subclustering process.

```R
# Load your ontology
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo",
                                  extract_tags = "everything")
# Run subclustering (only for cell types with descendants)
results <- subcluster_and_find_markers(
  seurat_obj = my_seurat,
  cl = cl,
  predicted_celltype_column = "celltype_parent",
  resolutions = c(0.1, 0.2, 0.3),
  reduction = "pca"
)
```

--------------------------------

### Calculate Pairwise Distances Example

Source: https://github.com/yrsong001/gptanno/blob/master/man/get_cl_pairwise_distances.Rd

Demonstrates calculating distances between two specific Cell Ontology terms using a graph object.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo")
graph <- build_ontology_graph(cl)
pairs <- data.frame(from = "CL:0000066", to = "CL:0000312")
distances <- get_cl_pairwise_distances(pairs, graph)
print(distances)
```

--------------------------------

### Search Cell Ontology terms using search_ols

Source: https://github.com/yrsong001/gptanno/blob/master/man/search_ols.Rd

Examples demonstrating basic queries, increasing result size for ambiguous terms, and automatic synonym matching.

```R
# Search for a cell type name
search_ols("fibroblast")

# Get more suggestions for ambiguous cell types
search_ols("helper T", size = 5)

# Automatic synonym matching
search_ols("ventricular cardiomyocyte")  # Returns "ventricular cardiac muscle cell"
```

--------------------------------

### Retrieve common ancestors for cell types

Source: https://github.com/yrsong001/gptanno/blob/master/man/get_common_ancestors.Rd

Demonstrates the setup of required ontology objects and the usage of get_common_ancestors to find shared ancestors between two pairs of cell types.

```R
# Setup required objects
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo",
                                  extract_tags = "everything")
cl_term_map <- build_cl_term_map(cl)
ancestor_type_map <- build_ancestor_type_map(cl)

# Get common ancestors
common <- get_common_ancestors("helper T cell", "cytotoxic T cell",
                              cl_term_map, cl, ancestor_type_map)
print(common)

# Check B cell vs plasma cell
common2 <- get_common_ancestors("B cell", "plasma cell",
                               cl_term_map, cl, ancestor_type_map)
print(common2)
```

--------------------------------

### Map cell types to CL IDs in R

Source: https://github.com/yrsong001/gptanno/blob/master/man/map_celltypes_to_cl.Rd

Example usage of the map_celltypes_to_cl function with a character vector of cell names.

```R
# cell_names <- c("fibroblast", "endothelial cell")
# map_celltypes_to_cl(cell_names, GPTAnno::cl_term_map)
```

--------------------------------

### Assign Subcluster Annotations to Seurat Object

Source: https://github.com/yrsong001/gptanno/blob/master/man/assign_subcluster_annotations_to_full.Rd

Examples demonstrating how to apply subcluster annotations to a full Seurat object using either automatic best-resolution selection or a manual resolution map.

```R
# Using auto-selected best resolutions
full_seurat <- assign_subcluster_annotations_to_full(
  full_seurat = my_seurat,
  workflow_results = workflow_results,
  use_best = TRUE,
  parent_column = "celltype_parent"
)
```

```R
# Using manual resolution selection
full_seurat <- assign_subcluster_annotations_to_full(
  full_seurat = my_seurat,
  workflow_results = workflow_results,
  use_best = FALSE,
  resolution_map = c("T cell" = "res_0.2", "B cell" = "res_0.1"),
  parent_column = "celltype_parent"
)
```

--------------------------------

### Configure LLM Backend

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Set up the LLM provider and model parameters. Supports OpenAI, Anthropic, and Gemini providers.

```r
Sys.setenv(OPENAI_API_KEY = "your-api-key")

llm_config <- list(
  provider = "openai",
  model = "gpt-5",
  params = ellmer::params(temperature = 0)
)
```

```r
llm_config <- list(provider = "anthropic", model = "claude-sonnet-4-20250514")
llm_config <- list(provider = "gemini", model = "gemini-2.0-flash")
```

--------------------------------

### Load GPTAnno Dependencies

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Initialize the required libraries for the GPTAnno workflow.

```r
library(GPTAnno)
library(Seurat)
library(ontologyIndex)
```

--------------------------------

### prepare_config()

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.html

Merges legacy model and llm_config parameters into a unified configuration object.

```APIDOC
## prepare_config(model, llm_config)

### Description
Utility to merge legacy model specifications and llm_config lists into a standard configuration format.

### Parameters
- **model** (string) - Optional - The model name.
- **llm_config** (list) - Optional - A list containing provider and model configuration details.
```

--------------------------------

### Initialize Cell Ontology Objects

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Sets up the ontology index and associated graph and mapping structures required for downstream analysis.

```r
library(ontologyIndex)

cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)
ancestor_type_map <- build_ancestor_type_map(cl)
cl_map <- build_cl_term_map(cl)
```

--------------------------------

### Configure OpenAI LLM Backend

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.html

Set the API key and configure the LLM provider for OpenAI.

```R
Sys.setenv(OPENAI_API_KEY = "your-api-key")

llm_config <- list(
  provider = "openai",
  model = "gpt-5",
  params = ellmer::params(temperature = 0)
)
```

--------------------------------

### Direct LLM Provider Interaction

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Use prepare_config to merge model settings and call_llm to execute prompts against a specified provider.

```r
cfg <- prepare_config(
  model = "gpt-5",
  llm_config = list(provider = "openai", model = "gpt-5")
)

call_llm(
  prompt = "Reply with the word fibroblast.",
  provider = cfg$provider,
  model = cfg$model
)
```

--------------------------------

### Initialize Ontology Graph for gptanno

Source: https://github.com/yrsong001/gptanno/blob/master/man/gptanno.Rd

Required steps to prepare the ontology object and graph structure before running the annotation workflow.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)
```

--------------------------------

### Configure API Credentials

Source: https://github.com/yrsong001/gptanno/blob/master/PDF2markers/README.md

Set the necessary environment variables for OpenAI API access.

```bash
export OPENAI_API_KEY="your-api-key"
export API_USER_ID="your-user-id"  # Optional
export API_PASSWORD="your-api-key"  # Alternative to OPENAI_API_KEY
```

--------------------------------

### Prepare LLM Configuration Usage

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_config.Rd

This function merges the model name and configuration list into a single structure for the chat client.

```R
prepare_config(model = "gpt-5.2", llm_config = NULL)
```

--------------------------------

### prepare_config(model = "gpt-5.2", llm_config = NULL)

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_config.Rd

Merges a model name and an optional configuration list into a unified LLM configuration structure.

```APIDOC
## prepare_config

### Description
Helper function that merges a legacy model parameter with a new llm_config structure. It provides backward compatibility while supporting new multi-provider configuration.

### Signature
`prepare_config(model = "gpt-5.2", llm_config = NULL)`

### Arguments
- **model** (Character) - Optional. Model name (default: "gpt-5.2"). Used if llm_config is not provided.
- **llm_config** (List) - Optional. Configuration containing: provider, model, params, temperature, max_tokens, api_key, api_url, system_prompt. If `params` is provided, it should be an `ellmer::params(...)` object.

### Returns
- **list** - A list containing the merged LLM configuration.
```

--------------------------------

### Configure Environment Variables

Source: https://github.com/yrsong001/gptanno/blob/master/PDF2markers/README.md

Sets required and optional environment variables for API authentication and model configuration.

```bash
# Required
export OPENAI_API_KEY="your-api-key"

# Optional
export API_USER_ID="your-user-id"
export API_PASSWORD="your-api-key"  # Alternative to OPENAI_API_KEY
export OPENAI_BASE_URL="https://api.openai.com/v1"  # Custom endpoint
export OPENAI_MODEL="gpt-4o-mini"  # Default model
```

--------------------------------

### Verify LLM Model Support

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

List available models or probe a specific model configuration for parameter compatibility.

```r
ellmer::models_openai()
```

```r
call_llm(
  prompt = "Reply with OK",
  provider = "openai",
  model = "gpt-5",
  params = ellmer::params(temperature = 0)
)
```

--------------------------------

### list_ollama_models(api_url = "http://localhost:11434")

Source: https://github.com/yrsong001/gptanno/blob/master/man/list_ollama_models.Rd

Lists all models currently available in Ollama. Requires Ollama to be running; returns a character vector of model names or NULL if inaccessible.

```APIDOC
## list_ollama_models

### Description
Lists all models currently available in Ollama. Requires Ollama to be running.

### Usage
list_ollama_models(api_url = "http://localhost:11434")

### Parameters
- **api_url** (Character) - Optional - Ollama API URL (default: "http://localhost:11434")

### Returns
- **Character vector** - A list of model names, or NULL if Ollama is not available.
```

--------------------------------

### Load Cell Ontology

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Initialize the ontology index and build the graph required for annotation.

```r
library(ontologyIndex)
cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)
```

--------------------------------

### Check Cell Ontology Relationship

Source: https://github.com/yrsong001/gptanno/blob/master/man/check_cl_relationship.Rd

Demonstrates how to retrieve the ontology, build the graph, and check the relationship between two CL IDs.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo")
graph <- build_ontology_graph(cl)
rel <- check_cl_relationship("CL:0000066", "CL:0000312", cl, graph)
print(rel)  # Should show child relationship
```

--------------------------------

### Prepare Marker Inheritance Strategy Configuration

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_marker_inheritance_strategy.Rd

Configures the marker inheritance strategy by specifying the parent marker file, the parent cell type, and the number of top genes to include.

```R
config <- prepare_marker_inheritance_strategy(
  parent_marker_file = "output/markers/markers_res_0.3.rds",
  parent_celltype = "T cell",
  topgenenumber = 10
)
```

--------------------------------

### Build Ontology Objects

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Download and process the Cell Ontology for use in downstream analysis.

```r
cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)
```

--------------------------------

### Load and inspect GPTCelltype_mapping

Source: https://github.com/yrsong001/gptanno/blob/master/man/GPTCelltype_mapping.Rd

Loads the mapping dataset and displays the first few entries.

```R
data(GPTCelltype_mapping)
head(GPTCelltype_mapping)
```

--------------------------------

### call_llm(prompt, provider, model, params, api_key, system_prompt, api_url)

Source: https://github.com/yrsong001/gptanno/blob/master/man/call_llm.Rd

Sends a prompt to a specified LLM provider and returns the generated text response.

```APIDOC
## call_llm

### Description
Routes prompts to OpenAI, Anthropic, Google Gemini, or local Ollama/vLLM models via the ellmer package.

### Parameters
- **prompt** (Character) - Required - The prompt to send.
- **provider** (Character) - Optional - One of "openai", "anthropic", "gemini", "ollama", or "vllm".
- **model** (Character) - Optional - Model name (e.g., "gpt-5.2", "claude-opus-4-6", "gemini-2.0-flash").
- **params** (Object) - Optional - An ellmer params object created by ellmer::params().
- **api_key** (Character) - Optional - API key for the provider. If NULL, uses environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, or GEMINI_API_KEY).
- **system_prompt** (Character) - Optional - System prompt to guide the model behavior.
- **api_url** (Character) - Optional - API URL for local LLMs (Ollama or vLLM). Defaults to "http://localhost:11434" for Ollama and "http://localhost:8000" for vLLM.

### Returns
- **Character** - The LLM response text.
```

--------------------------------

### Annotating Datasets with GPTAnno

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

R workflow for setting up ontology graphs, performing multi-resolution clustering, and running GPT-based cell annotation.

```r
# Download dataset (example)
# aging <- readRDS("path/to/Aging_preprocessed.rds")

# Setup
cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)

# Run clustering (use same path for gptanno marker_dir)
marker_dir <- "output/marker_genes"
cluster_results <- run_multi_resolution_clustering(
  seurat_obj = aging,
  resolutions = c(0.1, 0.3, 0.5, 0.7),
  result_dir = marker_dir
)
aging <- cluster_results$seurat_obj

# Perform repeated GPT queries for reference-free annotation
results <- gptanno(
  seurat_obj = aging,
  resolutions = c(0.1, 0.3, 0.5, 0.7),
  cl = cl,
  graph = graph,
  tissue_name = "mouse adult heart non-cardiomyocytes",
  llm_config = list(provider = "openai", model = "gpt-5"),  # or "gpt-4", etc.
  marker_dir = marker_dir,
  n_runs = 10  # Multiple queries enable reproducibility assessment
)
```

--------------------------------

### prepare_marker_inheritance_strategy(parent_marker_file, parent_celltype, user_restrict_to = NULL, topgenenumber = 10)

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_marker_inheritance_strategy.Rd

Configures the strategy for marker-inheritance based subcluster annotation.

```APIDOC
## prepare_marker_inheritance_strategy

### Description
Configures parameters for marker-inheritance based subcluster annotation. It combines parent cluster markers with subcluster markers to define the annotation strategy.

### Signature
`prepare_marker_inheritance_strategy(parent_marker_file, parent_celltype, user_restrict_to = NULL, topgenenumber = 10)`

### Parameters
- **parent_marker_file** (character) - Required - Path to the parent marker RDS file.
- **parent_celltype** (character) - Required - The name of the parent cell type.
- **user_restrict_to** (character vector) - Optional - User-specified cell type restrictions.
- **topgenenumber** (integer) - Optional - Number of top genes to include from each source (default: 10).

### Returns
- **List** - A list containing the strategy configuration.

### Example
```r
config <- prepare_marker_inheritance_strategy(
  parent_marker_file = "output/markers/markers_res_0.3.rds",
  parent_celltype = "T cell",
  topgenenumber = 10
)
```
```

--------------------------------

### Extract synonyms from a string

Source: https://github.com/yrsong001/gptanno/blob/master/man/extract_synonyms.Rd

Demonstrates basic usage of the extract_synonyms function with a simple character input.

```R
extract_synonyms("fibroblast")
```

--------------------------------

### build_cl_term_map(cl, verbose = TRUE)

Source: https://github.com/yrsong001/gptanno/blob/master/man/build_cl_term_map.Rd

Constructs a lookup table mapping CL names and synonyms to CL IDs from an ontology object.

```APIDOC
## build_cl_term_map

### Description
Constructs a lookup table mapping CL names and synonyms to CL IDs.

### Signature
`build_cl_term_map(cl, verbose = TRUE)`

### Parameters
- **cl** (ontologyIndex object) - Required - An ontologyIndex CL ontology object.
- **verbose** (logical) - Optional - Print mapping stats. Defaults to TRUE.

### Returns
- **data.frame** - A data frame with columns: key, clid, cl_label.
```

--------------------------------

### run_multi_resolution_clustering(seurat_obj, resolutions, result_dir, dims, assay, group.by, reduction, use_existing_neighbors)

Source: https://github.com/yrsong001/gptanno/blob/master/man/run_multi_resolution_clustering.Rd

Runs graph-based clustering at multiple resolutions and finds marker genes for each, saving results to disk.

```APIDOC
## run_multi_resolution_clustering

### Description
Runs graph-based clustering at multiple resolutions and finds marker genes for each, saving results to disk.

### Parameters
- **seurat_obj** (Seurat object) - Required - A Seurat object.
- **resolutions** (Numeric vector) - Required - Numeric vector of resolution parameters (e.g., c(0.1, 0.2, 0.3)).
- **result_dir** (Character) - Optional - Directory to save marker RDS files (default: "results/markers").
- **dims** (Numeric vector) - Optional - Which dimensions to use for clustering (default: 1:30).
- **assay** (Character) - Optional - Assay to use, or NULL for current (default: NULL).
- **group.by** (Character) - Optional - Metadata column to use for group assignment (default: "seurat_clusters").
- **reduction** (Character) - Optional - Reduction to use for graph construction (default: "pca").
- **use_existing_neighbors** (Logical) - Optional - Use an existing neighbor graph if present (default: TRUE).

### Returns
A list with: updated Seurat object, marker lists, and cluster assignments for each resolution.

### Example
```r
result <- run_multi_resolution_clustering(seurat_obj, c(0.1, 0.2, 0.3))
```
```

--------------------------------

### prepare_ontology_strategy(cl, parent_celltype, user_restrict_to = NULL, combine_restrictions = TRUE)

Source: https://github.com/yrsong001/gptanno/blob/master/man/prepare_ontology_strategy.Rd

Configures parameters for ontology-based subcluster annotation by determining the restriction list from ontology descendants and/or user input.

```APIDOC
## prepare_ontology_strategy

### Description
Configures parameters for ontology-based subcluster annotation. Determines the restriction list from ontology descendants, user input, or both.

### Arguments
- **cl** (ontology object) - Required. Ontology object from the ontologyIndex package.
- **parent_celltype** (character) - Required. The parent cell type name.
- **user_restrict_to** (character vector) - Optional. User-specified cell type restrictions.
- **combine_restrictions** (logical) - Optional. Controls how restrictions are determined. If TRUE (default) and user_restrict_to is provided, returns the union of ontology descendants and the user list. If FALSE and user_restrict_to is provided, returns only the user list. If user_restrict_to is NULL, returns ontology descendants.

### Returns
- **List** - Returns a list containing strategy configuration: strategy, restrict_to, parent_celltype, tissue_context, and marker_prep_fn.

### Example
```r
# Combine descendants with user list
config <- prepare_ontology_strategy(
  cl, "T cell",
  user_restrict_to = c("regulatory T cell", "gamma-delta T cell"),
  combine_restrictions = TRUE
)
```
```

--------------------------------

### Preprocess Seurat Object

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Run this optional step to normalize and prepare a Seurat object if it lacks PCA or UMAP data.

```r
seurat_obj <- preprocess_seurat_object(seurat_obj, npcs = 30, save_path = NULL)  # or save_path = "path/to/preprocessed.rds"
```

--------------------------------

### Load and inspect cl_term_map

Source: https://github.com/yrsong001/gptanno/blob/master/man/cl_term_map.Rd

Load the dataset into the R environment and display the first few rows to verify the structure.

```R
data(cl_term_map)
head(cl_term_map)
```

--------------------------------

### Constructing an igraph from an ontologyIndex object

Source: https://github.com/yrsong001/gptanno/blob/master/man/build_ontology_graph.Rd

Demonstrates the conversion of a Cell Ontology object into an igraph structure using parent relationships.

```R
# cl <- ontologyIndex::get_ontology(
#   "http://purl.obolibrary.org/obo/cl.obo",
#   extract_tags = "everything"
# )
# graph <- build_ontology_graph(cl)
```

--------------------------------

### Run subcluster annotation workflow

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Execute the annotation workflow using either an ontology-constrained or marker-inheritance strategy.

```r
subcluster_workflow_results <- run_subcluster_annotation_workflow(
  base_dir = "output/subclusters",
  strategy = "ontology",
  cl = cl,
  tissue_name = "mouse adult heart",
  resolutions = c(0.1, 0.2, 0.3),
  llm_config = llm_config,
  n_runs = 10,
  select_best = TRUE
)
```

```r
subcluster_workflow_results <- run_subcluster_annotation_workflow(
  base_dir = "output/subclusters",
  strategy = "marker_inheritance",
  cl = cl,
  parent_marker_root = marker_dir,
  parent_res = res_val,
  parent_cluster_col = cluster_col,
  tissue_name = "mouse adult heart",
  llm_config = llm_config,
  n_runs = 10
)
```

--------------------------------

### Run Extraction Pipeline

Source: https://github.com/yrsong001/gptanno/blob/master/PDF2markers/README.md

Execute the extraction script in interactive, direct, or advanced modes.

```bash
python paper_extraction_cellNgenes.py
```

```bash
python paper_extraction_cellNgenes.py --pdf "path/to/your/paper.pdf" --out "outputs"
```

```bash
python paper_extraction_cellNgenes.py \
    --pdf "papers/example.pdf" \
    --out "outputs" \
    --model "gpt-4o-mini" \
    --limit 100
```

--------------------------------

### Configure Ollama in R

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Define the llm_config list to point to the local Ollama instance for use in annotation workflows.

```r
llm_config = list(
  provider = "ollama",
  model = "llama3",
  api_url = "http://localhost:11434"
)
# Use the same llm_config in gptanno(), run_subcluster_annotation_workflow(), etc.
```

--------------------------------

### Select Best Resolution

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Extracts the best resolution name and corresponding object based on the composite score.

```r
best_res_name <- scores$resolution[1]
best_res_obj <- annotation_results[[best_res_name]]
res_val <- sub("^res_", "", best_res_name)
cluster_col <- paste0("cluster_res.", res_val)
```

--------------------------------

### Run GPTAnno Workflow

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Perform clustering, marker gene identification, and GPT-based annotation.

```r
library(GPTAnno)
library(Seurat)
library(ontologyIndex)
library(dplyr)

# Setup
Sys.setenv(OPENAI_API_KEY = "your-api-key")
cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)

# Run clustering and find markers (writes markers to result_dir)
marker_dir <- "output/marker_genes"
cluster_results <- run_multi_resolution_clustering(
  seurat_obj = your_seurat_object,
  resolutions = c(0.1, 0.3, 0.5),
  result_dir = marker_dir
)
your_seurat_object <- cluster_results$seurat_obj

# Run repeated GPT queries at multiple resolutions (reads markers from marker_dir)
results <- gptanno(
  seurat_obj = your_seurat_object,
  resolutions = c(0.1, 0.3, 0.5),
  cl = cl,
  graph = graph,
  tissue_name = "mouse heart",
  llm_config = list(
    provider = "openai",
    model = "gpt-5" # Optionally can change params, eg. params = ellmer::params(temperature = 1)
  ),
  marker_dir = marker_dir,
  n_runs = 2  # Number of independent queries for reproducibility assessment
)

# Evaluate resolutions with composite scoring
scores <- score_annotation_resolutions(results)
print(scores)  # Highest composite score is optimal resolution
```

--------------------------------

### call_llm()

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.html

Executes a direct call to an LLM provider.

```APIDOC
## call_llm(prompt, provider, model)

### Description
Directly invokes an LLM provider with a specified prompt and model configuration.

### Parameters
- **prompt** (string) - Required - The text prompt to send to the LLM.
- **provider** (string) - Required - The LLM provider name.
- **model** (string) - Required - The specific model identifier.
```

--------------------------------

### summarize_gptcelltype(markers, model = "gpt-5", tissue_name = "", n_runs = 2, topgenenumber = 10, add_cl_prompt = FALSE, restrict_to = NULL, parent_celltype = NULL, llm_config = NULL)

Source: https://github.com/yrsong001/gptanno/blob/master/man/summarize_gptcelltype.Rd

Calls an LLM multiple times for cell type annotation and summarizes the results across runs.

```APIDOC
## summarize_gptcelltype

### Description
Calls LLM multiple times for cell type annotation and summarizes results.

### Parameters
- **markers** (Named character vector or list) - Required - Marker genes per cluster.
- **model** (Character) - Optional - Model to use (default: 'gpt-5').
- **tissue_name** (Character) - Optional - Context for prompt.
- **n_runs** (Integer) - Optional - Number of LLM calls to aggregate (default: 2).
- **topgenenumber** (Integer) - Optional - Number of top marker genes per cluster (default: 10).
- **add_cl_prompt** (Logical) - Optional - If TRUE, adds Cell Ontology request to prompt.
- **restrict_to** (Character vector) - Optional - Restrict predictions to these cell types.
- **parent_celltype** (Character) - Optional - Predict child cell type of this parent.
- **llm_config** (List) - Optional - LLM configuration passed to gptcelltype (supports OpenAI, Anthropic, and Google Gemini).

### Returns
- **combined_results** (List) - Raw predictions from successful runs.
- **summary** (Table) - Weighted summary table.
- **final_summary** (Table) - Most frequent annotation per cluster.
- **run_summary** (Table) - Failed runs with note column.
```

--------------------------------

### check_ollama_available(api_url = "http://localhost:11434")

Source: https://github.com/yrsong001/gptanno/blob/master/man/check_ollama_available.Rd

Checks if the Ollama service is running and accessible at the provided API URL.

```APIDOC
## check_ollama_available

### Description
Checks if Ollama is running and accessible at the specified URL. This function attempts to ping the Ollama API to verify connectivity.

### Signature
`check_ollama_available(api_url = "http://localhost:11434")`

### Parameters
- **api_url** (Character) - Optional - Ollama API URL (default: "http://localhost:11434")

### Returns
- **Logical** - TRUE if Ollama is accessible, FALSE otherwise.

### Example
```r
check_ollama_available()
```
```

--------------------------------

### Run Multi-Resolution Clustering

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Perform clustering across multiple resolutions and save marker gene results to the specified directory.

```r
marker_dir <- "output/marker_genes"

cluster_results <- run_multi_resolution_clustering(
  seurat_obj = seurat_obj,
  resolutions = c(0.1, 0.5, 0.9, 1.2),
  result_dir = marker_dir,
  dims = 1:30
)

seurat_obj <- cluster_results$seurat_obj
```

--------------------------------

### run_subcluster_annotation_workflow(base_dir, strategy, cl, ...)

Source: https://github.com/yrsong001/gptanno/blob/master/man/run_subcluster_annotation_workflow.Rd

Executes the subcluster annotation workflow using specified strategies and cell-type restrictions.

```APIDOC
## run_subcluster_annotation_workflow

### Description
Executes a subcluster annotation workflow. This function allows users to define annotation strategies (ontology or marker_inheritance) and apply specific cell-type restrictions to refine the annotation process.

### Parameters
- **base_dir** (string) - Required - Directory path for subcluster output.
- **strategy** (string) - Required - Annotation strategy to use (e.g., "ontology", "marker_inheritance").
- **cl** (object) - Required - Cluster information object.
- **user_restrict_to** (list/vector) - Optional - Cell-type restrictions to apply.
- **combine_restrictions** (boolean) - Optional - Whether to combine user restrictions with ontology descendants.
- **tissue_name** (string) - Optional - Name of the tissue being analyzed.
- **select_best** (boolean) - Optional - Whether to select the best annotation result.
- **celltypes_to_subcluster** (string) - Optional - Specific cell type to target for subclustering.
- **parent_marker_root** (string) - Optional - Path to parent marker files (required for marker_inheritance).
- **parent_res** (string) - Optional - Resolution of the parent clustering.
- **parent_cluster_col** (string) - Optional - Column name for parent cluster identifiers.
```

--------------------------------

### Run Ontology-based Post-processing

Source: https://github.com/yrsong001/gptanno/blob/master/PDF2markers/README.md

Executes the filterout_cell_ontology script to process CSV files in the specified directory.

```bash
python filterout_cell_ontology.py --input-dir ./outputs
```

--------------------------------

### Standardize and Match Annotations

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Standardizes raw cell labels and attempts to map them to known Cell Ontology naming conventions.

```R
library(GPTAnno)

clean_and_match_annotation("Helper T cells")
clean_and_match_annotation("cardiomyocytes")
clean_and_match_annotation("unknown cell")
```

--------------------------------

### Multi-Resolution Clustering and Annotation

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Perform adaptive clustering and use GPT to annotate cell types based on ontology and marker genes.

```r
# Setup ontology
cl <- get_ontology("http://purl.obolibrary.org/obo/cl.obo", extract_tags = "everything")
graph <- build_ontology_graph(cl)

# Step 1a: Run clustering and find markers (use same path for Step 1b marker_dir)
marker_dir <- "output/marker_genes"
cluster_results <- run_multi_resolution_clustering(
  seurat_obj = seurat_obj,
  resolutions = c(0.1, 0.3, 0.5, 0.7),
  result_dir = marker_dir,
  dims = 1:30
)
seurat_obj <- cluster_results$seurat_obj

# Step 1b: Perform repeated GPT queries for each resolution
parent_results <- gptanno(
  seurat_obj = seurat_obj,
  resolutions = c(0.1, 0.3, 0.5, 0.7),
  cl = cl,
  graph = graph,
  tissue_name = "mouse adult heart",
  llm_config = list(provider = "openai", model = "gpt-5"),  # or "gpt-4", etc.
  marker_dir = marker_dir,
  n_runs = 10,  # Repeated queries for reproducibility measures
  topgenenumber = 10
)

# Step 1c: Score resolutions using composite framework
scores <- score_annotation_resolutions(parent_results)
print(scores)

# Step 1d: Select optimal resolution and assign annotations
best_res_name <- scores$resolution[1]  # e.g., "res_0.3"
best_res_obj <- parent_results[[best_res_name]]
res_val <- sub("^res_", "", best_res_name)  # Extract "0.3"
cluster_col <- paste0("cluster_res.", res_val)  # "cluster_res.0.3"

seurat_obj <- assign_celltype(
  seurat_obj = seurat_obj,
  annotation_summary = best_res_obj,
  cluster_col = cluster_col,
  new_celltype = "celltype_parent"
)
```

--------------------------------

### Run Marker-Inheritance Subcluster Annotation

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.html

Use this strategy for discovery-oriented subtype annotation.

```R
subcluster_workflow_results <- run_subcluster_annotation_workflow(
  base_dir = "output/subclusters",
  strategy = "marker_inheritance",
  cl = cl,
  parent_marker_root = marker_dir,
  parent_res = res_val,
  parent_cluster_col = cluster_col,
  tissue_name = "mouse adult heart",
  llm_config = llm_config,
  n_runs = 10
)
```

--------------------------------

### Evaluate annotation agreement with ontology

Source: https://github.com/yrsong001/gptanno/blob/master/README.md

Build an ancestor map and score agreement between manual and predicted annotations.

```r
ancestor_type_map <- build_ancestor_type_map(cl)
agreement <- score_annotation_agreement_ontology(
  seurat_obj,
  manual_col = "celltype_manual",
  predicted_col = "celltype_parent",
  ancestor_type_map = ancestor_type_map
)
print(agreement$summary)
```

--------------------------------

### Retrieve direct parent terms for a CL ID

Source: https://github.com/yrsong001/gptanno/blob/master/man/get_cl_direct_parents.Rd

Demonstrates how to load the Cell Ontology and fetch direct parents for a specific CL identifier.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo")
parents <- get_cl_direct_parents("CL:0000066", cl)  # epithelial cell parents
print(parents) # CL:0000255 eukaryotic cell
```

--------------------------------

### build_ontology_graph(ontology)

Source: https://github.com/yrsong001/gptanno/blob/master/man/build_ontology_graph.Rd

Constructs an igraph directed graph from an ontologyIndex object using 'is_a' parent relationships.

```APIDOC
## build_ontology_graph

### Description
Constructs an igraph directed graph from an ontologyIndex object using 'is_a' parent relationships. The resulting graph nodes are named by their ontology term IDs.

### Signature
`build_ontology_graph(ontology)`

### Parameters
- **ontology** (ontologyIndex object) - Required - An ontologyIndex object containing a $parents list (e.g., as returned by ontologyIndex::get_ontology()).

### Returns
- **igraph** (igraph object) - A directed graph representation of the ontology.

### Example
```r
# cl <- ontologyIndex::get_ontology(
#   "http://purl.obolibrary.org/obo/cl.obo",
#   extract_tags = "everything"
# )
# graph <- build_ontology_graph(cl)
```
```

--------------------------------

### Preprocess Seurat Object

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Prepare the Seurat object for clustering if it has not been normalized or reduced.

```r
seurat_obj <- preprocess_seurat_object(
  seurat_obj,
  npcs = 30
)
```

--------------------------------

### Check Ollama Availability

Source: https://github.com/yrsong001/gptanno/blob/master/man/check_ollama_available.Rd

Verifies if the Ollama service is reachable at the default or specified API URL.

```R
check_ollama_available()
```

--------------------------------

### assign_subcluster_annotations_to_full(full_seurat, workflow_results, use_best = TRUE, resolution_map = NULL, parent_column = "celltype_parent", final_colname = "celltype_subcluster", annotation_col_pattern = "^annotated_sub_", clean_labels = TRUE)

Source: https://github.com/yrsong001/gptanno/blob/master/man/assign_subcluster_annotations_to_full.Rd

Assigns subcluster annotations to a full Seurat object based on results from the subcluster annotation workflow.

```APIDOC
## assign_subcluster_annotations_to_full

### Description
Unified function that works for both ontology and marker-inheritance strategies. Takes subcluster annotation results and assigns them back to the full Seurat object.

### Parameters
- **full_seurat** (Seurat object) - Required - Full Seurat object (not subset).
- **workflow_results** (List) - Required - List returned by run_subcluster_annotation_workflow() containing results, scores, best, and strategy.
- **use_best** (Logical) - Optional - If TRUE, use best resolution per celltype. If FALSE, user must specify resolution_map. Default: TRUE.
- **resolution_map** (Named vector) - Optional - Specify resolution per celltype (e.g., c("T cell" = "res_0.1")).
- **parent_column** (Character) - Optional - Metadata column with parent cell type annotation. Default: "celltype_parent".
- **final_colname** (Character) - Optional - Name for new annotation column. Default: "celltype_subcluster".
- **annotation_col_pattern** (Character) - Optional - Pattern to find annotation column in subset. Default: "^annotated_sub_".
- **clean_labels** (Logical) - Optional - Remove " (not in GO)" suffix. Default: TRUE.

### Returns
- **Seurat object** - Updated full Seurat object with subcluster annotations.
```

--------------------------------

### Retrieve direct child terms for a CL ID

Source: https://github.com/yrsong001/gptanno/blob/master/man/get_cl_direct_children.Rd

Demonstrates how to load the Cell Ontology and retrieve direct children for a specific ID.

```R
cl <- ontologyIndex::get_ontology("http://purl.obolibrary.org/obo/cl.obo")
children <- get_cl_direct_children("CL:0000066", cl)  # epithelial cell children
print(children)
```

--------------------------------

### Run LLM Annotation Across Resolutions

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/gptanno-pipeline.Rmd

Executes repeated LLM annotations for specified clustering resolutions and saves the results to an RDS file.

```r
annotation_results <- gptanno(
  seurat_obj = seurat_obj,
  resolutions = c(0.1, 0.5, 0.9, 1.2),
  cl = cl,
  graph = graph,
  tissue_name = "mouse adult heart",
  llm_config = llm_config,
  marker_dir = marker_dir,
  n_runs = 10,
  topgenenumber = 10
)

saveRDS(annotation_results, "annotation_results.rds")
```

--------------------------------

### Inspect Ontology Hierarchy

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Lower-level helpers for retrieving parents, children, node depths, pairwise distances, and names for CL IDs.

```r
get_cl_direct_parents("CL:0000067", cl)
get_cl_direct_children("CL:0000115", cl)
get_cl_node_depths(c("CL:0000067", "CL:0000115"), graph)
get_cl_pairwise_distances(
  data.frame(from = "CL:0000067", to = "CL:0000115"),
  graph
)
get_cl_names(c("CL:0000067", "CL:0000115"), cl)
```

--------------------------------

### Run Subcluster Annotation with Marker Inheritance

Source: https://github.com/yrsong001/gptanno/blob/master/man/run_subcluster_annotation_workflow.Rd

Utilize the marker inheritance strategy by providing parent cluster information and specific cell-type restrictions.

```R
workflow_results <- run_subcluster_annotation_workflow(
  base_dir = "output/subclusters",
  strategy = "marker_inheritance",
  cl = cl,  # Required for ontology distance calculation
  parent_marker_root = "output/markers",
  parent_res = "0.3",
  parent_cluster_col = "cluster_res.0.3",
  user_restrict_to = list(
    "T cell" = c("Treg", "Helper T cells", "cytotoxic T cell")
  ),
  tissue_name = "mouse heart"
)
```

--------------------------------

### clean_and_match_annotation(annotation, use_ols = FALSE)

Source: https://github.com/yrsong001/gptanno/blob/master/man/clean_and_match_annotation.Rd

Cleans, standardizes, and maps raw cell type predictions to Cell Ontology names.

```APIDOC
## clean_and_match_annotation

### Description
Cleans, standardizes, and maps raw cell type predictions to Cell Ontology names using GPTCelltype_mapping and cl_term_map.

### Signature
`clean_and_match_annotation(annotation, use_ols = FALSE)`

### Parameters
- **annotation** (Character) - Required - Raw annotation string (possibly mixed or ambiguous).
- **use_ols** (Logical) - Optional - Whether to use OLS search as fallback (default: FALSE). Please use this with Caution.

### Returns
- **Character** - Cleaned and mapped annotation string, or the original in lower singular form if not matched.

### Examples
```r
clean_and_match_annotation("t cell")
clean_and_match_annotation("Helper T cells")
clean_and_match_annotation("unknown cell")
```
```

--------------------------------

### List Ollama Models

Source: https://github.com/yrsong001/gptanno/blob/master/man/list_ollama_models.Rd

Retrieves the list of available models from the default Ollama API endpoint.

```R
list_ollama_models()
```

--------------------------------

### map_celltypes_to_cl(terms, cl_term_map = GPTAnno::cl_term_map, verbose = TRUE)

Source: https://github.com/yrsong001/gptanno/blob/master/man/map_celltypes_to_cl.Rd

Maps a character vector of cell type names to CL ontology IDs using a prebuilt mapping table.

```APIDOC
## map_celltypes_to_cl

### Description
Maps a character vector of cell type names to CL ontology IDs using a prebuilt mapping table.

### Parameters
- **terms** (Character vector) - Required - Character vector of cell type names.
- **cl_term_map** (Data.frame) - Optional - Defaults to package built-in map, or returned by build_cl_term_map().
- **verbose** (Logical) - Optional - Print mapping results.

### Returns
- **Data.frame** - Returns a data frame with columns: term, clid, cl_label.
```

--------------------------------

### Check Cell Type Relationships

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Compares two CL IDs to determine their hierarchical relationship.

```r
check_cl_relationship("CL:0000066", "CL:0000312", cl, graph)
```

--------------------------------

### Score Annotation Resolutions

Source: https://github.com/yrsong001/gptanno/blob/master/vignettes/helper-utilities.Rmd

Computes a composite score to rank annotation strategies.

```r
score_annotation_resolutions(annotation_result_list)
```