UMass Chan Bioinformatics CoreBioinformatics Core
GitHub

Bootcamp / Session 1 / 1.3

Session 1.3 · Act two · Ming “Tommy” Tang, AstraZeneca

AI in bioinformatics: will you be the pilot or the passenger?

What an AI coding agent is, how a working bioinformatician uses one from plan to published analysis, why the foundations matter more than ever, and a live Claude Code demo on an RNA-seq analysis.

Download the slides Live demo workshop

Recording

Tommy Tang · 53 minutes · Download (117 MB)

What the talk covers

  1. Chat versus agent. How an AI coding agent differs from ChatGPT or Claude chat: it reads files, runs commands, and iterates.
  2. Leverage. A Shiny app in a few days, a personal website, a thinking partner, and the analysis itself.
  3. Foundations still matter. Unix, R or Python, and biology. Your experience is what spots the batch effect in a PCA.
  4. A real workflow. Plan first, point the agent at your old scripts, generate an Rmd or Jupyter notebook, run it line by line, commit to GitHub, interpret with biologists.
  5. Risks. AI analysis is not reproducible, traceability is a different question, and hallucinated data is the real danger.
  6. Claude Code in practice. Permission modes, thinking effort, model choice, context management, subagents, CLAUDE.md, plugins, and skills.

Commands from the talk

CommandWhat it does
/planPlan before acting. Give it the full context: slides, PDFs, data, links.
/initCreate a CLAUDE.md that documents the project and its instructions.
/modelPick the model that fits the task.
/effortSet how hard it thinks for the whole session. “ultrathink” raises it for one message.
/compactSummarize a long session, optionally with instructions on what to keep.
/rewindGo back to an earlier point in the session.
claude --resumePick up a previous session.
claude --permission-mode autoLet it act without asking each time. Fast; see Part 1.4 for when not to.
/pluginsInstall plugins such as Superpowers and frontend-design, which bundle skills.

Key takeaways: embrace AI, master the foundations, and remember that AI is probabilistic, so verify the results.

Materials

SlidesAI in Bioinformatics (PowerPoint)The deck presented in class, 35 slides.GitHub · session1/ SlidesLive demo slidesTommy's Google Slides for the Claude Code demo.docs.google.com WorkshopClaude Code RNA-seq workshopThe step-by-step analysis from the live demo.umms-biocore.github.io BlogDiving into Genetics and GenomicsTommy's blog, including the PCA and batch-effect posts.divingintogeneticsandgenomics.com VideoChatomics on YouTubeBioinformatics tutorials and talks.youtube.com/@chatomics