Create a Blueprint using Natural Language — inquire

Ask a Large Language Model (LLM) to create a blueprint for glycomics or glycoproteomics data analysis. To use this function, you need to have a DeepSeek API key. You can get a DeepSeek API key from https://platform.deepseek.com. Then set the environment variable DEEPSEEK_API_KEY to your API key with Sys.setenv(DEEPSEEK_API_KEY = "your-api-key").

Usage

inquire_blueprint(
  description,
  exp = NULL,
  group_col = "group",
  model = "deepseek-chat",
  max_retries = 3
)

Arguments

description: A description of what you want to analysis.
exp: Optional. A glyexp::experiment() object to provide more context to the LLM.
group_col: The column name of the group variable in the experiment. Default to "group".
model: Model to use. Default to "deepseek-chat".
max_retries: Maximum number of retries when the AI output is invalid. Default to 3.

Details

LLMs can be unstable. If you get an error, try again with another description. Make sure to examine the returned blueprint carefully to ensure it's what you want. You can also create parallel analysis branches with br("name", step_..., step_...), which will namespace outputs with the branch prefix. If the LLM needs required information to proceed, it may ask clarifying questions interactively and then retry with your answers. After a blueprint is generated, the description is printed and, in interactive sessions, you can press ENTER to accept it or type new requirements to refine the blueprint. This review step can repeat until you accept the plan.

Here are some examples that works:

"I want to know what pathways are enriched for my differentially expressed glycoforms."
"I want a heatmap and a pca plot. I have already performed preprocessing myself."
"I have a glycomics dataset. I want to calculate derived traits and perform DEA on them."