AI-Enhanced Bioinformatics - MUNI Summer School 2026

June 25, 2026 · View on GitHub

A four-week online prep course followed by a one-week in-person course on agentic development for bioinformaticians and life-science students. The prep teaches enough about LLMs, agentic IDEs, bio foundation models, and the agent–tool ecosystem that students arrive in Brno ready to build, not to be lectured at.

Hosted by Masaryk University. Course page: https://summeratmasaryk.cz/aienhanced-bioinformatics-educ

Schedule at a glance

WeekDates (Mon–Fri)ThemeLive sessions (Europe/Prague)
125 May – 29 May 2026What an LLM actually isMon 25.5 · 09:00–10:00 kick-off
Tue 26.5 · 11:00–14:00 Domain Expertise: The Internal Validator of AI Quality
Fri 29.5 · 12:00–12:30 Q&A
21 Jun – 5 Jun 2026Agentic IDEs on bioinformatics tasksTue 2.6 · 11:00–14:00 Feature Engineering: Translating Scientific Intuition into Numbers
Fri 5.6 · 12:00–12:30 Q&A
38 Jun – 12 Jun 2026Bio foundation models as toolsTue 9.6 · 11:00–14:00 Interpretability: Translating AI Outputs Back into Human Knowledge
Fri 12.6 · 12:00–12:30 Q&A
415 Jun – 19 Jun 2026How agents talk to the world: tools, commands, MCPFri 19.6 · 12:00–12:30 Q&A
In person22 Jun – 26 Jun 2026Build week, Brno (Czechia)Wed 24.6 · 09:00–13:00 From Domain Knowledge to Working Models
Thu 25.6 · 09:00–13:00 The Power of Autoencoders: Human–Machine Translators
Fri 26.6 · 09:00–12:00 From AI Outputs to Scientific Judgment: Strengths, Limits, and Failure Modes

All Q&A slots are 30 minutes, drop-in, no prepared talk. Bring questions, broken pipelines, and confused agents.

Course philosophy

Three principles run through every week:

  1. Build alongside the agent, not under it. Every exercise produces a real artifact - a script, a notebook, a small repo - that ends up in version control. Nothing is throwaway.
  2. Biological invariants are the cheapest validator you have. Agents will produce code that runs, prints plausible output, and is silently wrong. The single most transferable skill in the course is the habit of running domain-aware checks on agent-generated artifacts. This thread is explicit from week 2 onward.
  3. Lessons accumulate. Across all four weeks, you keep one lessons.md in your repo with two kinds of entry each week: From the materials (short video/paper notes and reflection exercises) and Surprises (concrete LLM and agent moments - tool, prompt, outcome). By week 4 this file is the most valuable artifact of the prep period.

The running project

Across the four weeks, you build one cohesive deliverable: an agentic pipeline that touches real bioinformatics tools and a real bio foundation model, lives in a Git repo, and is reproducible by someone else. Each week adds a layer:

  • Week 1. First chatbot interactions, GitHub and Slack accounts, this repo forked or cloned, lessons.md started.
  • Week 2. Agent-generated bioinformatics code in the repo, with versioned commits and real tests against the trap exercise.
  • Week 3. Foundation-model evaluation added - a genomic LM benchmarked against the published baseline on Genomic Benchmarks.
  • Week 4. Reproducibility scaffolding (pyproject.toml, AGENTS.md), MCP demo, tagged release. Ready to bring to the in-person week.

Day one of the in-person week, you'll bring this repo plus lessons.md plus a list of three things you want to build.

Tools and platforms

  • IDE: Antigravity (free for individuals and students). You can use Cursor and/or Claude Code if you are experienced with them.
  • Notebooks: Google Colab for compute-heavy exercises (free tier sufficient for everything in the course).
  • Version control and storage: GitHub. Everything ends up there.
  • Cohort chat: Slack. Create an account in week 1 (see email); day-to-day questions, links, and logistics go there (workspace invite in the kick-off email).
  • Compute constraints assumed: Colab free tier (T4 when available, CPU otherwise). All exercises sized to fit.

How to engage with the videos and papers

Watching is not learning. For each video, stop every ~20 minutes and add a line under that week's From the materials section in lessons.md: "what's the one thing I'd want to test from what I just heard?" For each paper, the same: "what claim would I most want to verify on my own data?"

LLM and agent surprises go under Surprises in the same file - with enough detail to be useful later (tool, prompt, what came back).

A short ungraded Google Form quiz per week (~5 application-style questions, answers revealed afterward) gives a soft accountability signal without creating evaluation overhead. Quizzes are open-everything; questions are designed so an LLM alone can't answer them - they require integration with your own context.

Assessment

Three things, that's it:

  1. A working repo (this one, forked or cloned) with weekly commits.
  2. lessons.md - both subsections each week filled in as you go (From the materials + Surprises).
  3. A one-paragraph reflection at the end of week 4: "What would I trust an agent to do in my research, and what would I not?"

Anyone who can answer that reflection concretely after four weeks has gotten what the prep was meant to deliver.

In-person week (22–26 June 2026, Brno)

The in-person week shapes the prep. Day one in person: pair up, exchange repos, and try to reproduce each other's week 3 results. That single arrangement raises the quality of every week 4 deliverable, so plan for it.

Repository layout

.
├── README.md            # this file
├── weeks/
│   ├── week1.md         # What an LLM actually is
│   ├── week2.md         # Agentic IDEs on bioinformatics tasks
│   ├── week3.md         # Bio foundation models as tools
│   ├── week4.md         # Tools, commands, MCP
│   └── cheatsheets/
│       └── bio-fm.md    # Week 3 model map
├── exercises/
│   ├── week2/
│   │   ├── broken_script.py
│   │   ├── example.fa
│   │   └── trap/
│   │       ├── genome.fa
│   │       ├── annotations.gff3
│   │       └── README.md
│   └── week3/
│       ├── README.md
│       ├── fetch_proteins.py
│       ├── protein_accessions.tsv
│       ├── proteins.fasta
│       └── results.md
├── onsite/              # in-person build week (placeholder schedule)
├── lessons.md           # prep log (From the materials + Surprises); template included
└── LICENSE

The exercises/ directory contains small starter assets for the hands-on weeks. Students add their own scripts, notebooks, results, and mini-project files as they go.

Contact

  • Day-to-day: cohort Slack workspace (invite in the kick-off email)
  • General contact: Mai Hoa Magnus - +420 775 853 057 - magnus@czs.muni.cz

Issues with the repo or course materials: open a GitHub issue against this repository.

License

Course materials in this repository are released under the Apache License 2.0. Use, fork, and adapt freely.