-
Notifications
You must be signed in to change notification settings - Fork 1
Prompting tips
Current as of 20 Feb 2026
- Ask things step by step to avoid overwhelming the agent.
- Suppose you want to find all genes that are differentially expressed in patients treated with VRD.
- Option 1: One-shot prompt
"Find all genes that are differentially expressed in patients treated with VRD."
- Realistically, the agent will most likely crash as it can only think 50 steps.
- Option 2: Divide-and-conquer
- "Find the list of patients treated with VRD (VRD), and the list of patients treated with other therapies (OTHR)."
- "Suppose my gene of interest is NFkB. Perform Wilcox rank sum test on the log2tpm expression values between VRD and OTHR.
- "Save the test results to VRD-OTHR-NFkB.csv"
- "Now I want to perform wilcox rank sum test on all genes. Do it one gene at a time. In each step, save the results to VRD-OTHR-[XXX].csv, where [XXX] is the gene symbol.
- Do you see the difference? The agent is more likely to return something meaningful in the second style of prompting.
-
Prompt: "save the results to a file and generate a link to download it"
-
Sometimes, the file is already saved but to a random path e.g.,
result/result_6e8571cf.csv. You can see the path in the tool result or AI response. -
To access this file, open a new browser tab and navigate to "https://chat.myegpt.com/result/result_6e8571cf.csv".
-
Alternatively, prompt "save the results file as [your-desired-result-path.csv] and generate a download link"
-
Suppose you want to make a publication-quality plot. You can prompt MyeGPT to iteratively adjust the plot to your needs.
-
The plotting engine is Pyplot from matplotlib. Official Pyplot Documentation
-
Typical prompts include "Re-plot and increase DPI to 300", "Re-plot and save figure as 8 by 6 inches", "Re-plot using [my-desired-color-map]" [key word: re-plot]
-
Refer to the anatomy of matplotlib figure below, to know which terms to use:
-
E.g., "Re-plot with larger axis tick labels, legend position on bottom-left"
-
Refer to the names of the matplotlib color maps below: