Using a generative AI to implement a script for bioanalysis

This folder contains the R implementation of the execise realised during the workshop “Using generative AI to implement a script for bioanalysis”, organised in Paris on June 13, 2025.

Source article: Kelliher, C. M., Leman, A. R., Sierra, C. S. & Haase, S. B. Investigating Conservation of the Cell-Cycle-Regulated Transcriptional Program in the Fungal Pathogen, Cryptococcus neoformans. PLoS Genet 12, e1006453 (2016). DOI 10.1371/journal.pgen.1006453

Pre-processing

Performed by Gaëlle Lelandais and the students of the “Diplôme Universitaire Omique (DUO)”

• The data table contains normalised counts of the 1705 yeast genes showing oscillating profiles of transcription during the cell cycle.
• Raw RNA-seq counts were normalised uusing R package DeSEQ2

Goal of the exercise

Use an AI to generate an R script that performs the following tasks:

• Download a data table from this Web site, and save it to a local folder
• Load this data table data in R
• Compute summary statistics on the whole data table (min, max, median, mean, sd, …)
• Compute marginal statistics on rows (genes) and columns (timepoints) of the data table

• Normalise the data table in 3 steps

  • log2 transformation
  • column-wise standardisation (centering and scaling)
  • row-wise standardisation (centering and scaling)
• Produce a heatmap similar to the leftv panel of Figure 2 of the source article (Kelliher, 2016)
• Export this heatmap in png and pdf formats

Some recommendations

• Provide your AI with context from the outset: tell it your goal, give it details of the type of data, your criteria (documentation, structuring, etc.)
• Act step-by-step: rather than asking for a script that does everything from the outset, start by asking for a script that does the first step (downloading the data) then the second (loading the data), and advance step by step.
• Structure the script: ask your AI to segment the code by implementing functions for each of the different tasks.
• Perform unit tests: design a validation test for each function and ask the AI to check that the code gives the right answer.

Results

The ChatGPT prompting is available here: yeast-cell-cycle_ChatGPT-prompting.md

The final code is available here : yeast-cell-cycle_analysis_2025-06-13.R