← Back to assessment

Project 2 · 70% · Week 11

Does your gene respond to p53 loss?

One gene, one cancer type, real patient data from The Cancer Genome Atlas. Decide whether that gene’s expression responds to loss of p53 function in human tumours — and if so, in which direction.

Why this is a real question

TP53 is the most frequently mutated gene in human cancer. Its protein, p53, is a transcription factor: when a cell is damaged, it switches on genes that stop the cell cycle or trigger cell death. So when p53 is lost, a gene can go either way:

Direct p53 target

goes down

the switch is gone

Proliferation gene

goes up

the brake is gone

Neither

no change

nothing to release

You cannot guess which your gene is — you have to measure it. “No detectable difference” is a real, valid, fully markable answer.

Your gene is assigned to you

Everyone has a different gene and cancer type. You never type them yourself: the setup cell at the top of each notebook asks for your 6-digit registration number — the number on Canvas and your student card — and looks yours up.

REGISTRATION_NUMBER = "123456"   # <-- yours

If the notebook says your number is not on the class list, check it against Canvas, then email the convenor. Do not pick a gene yourself — you are marked on the one assigned to you.

Get started — Colab or your own computer

Your choice; neither route earns or loses a mark. Start with 00_start_here.

Option 1 — Google Colab

Nothing to install.

  • Open in ColabStart hereThe question, how it is marked, and how to work.
  • Open in ColabNotebook 01 — the cohortClean a messy patient export and write your Table 1.
  • Open in ColabNotebook 02 — the analysisTest your gene, draw the figure, write your one-sentence answer.
  • Open in ColabNotebook 03 — the extension (optional)One deeper analysis of your choosing.
Open a notebook, then File → Save a copy in Drive before you type anything — otherwise your work is not saved.

Option 2 — on your own computer

Download the project and set it up the way the tech sessions showed you:

curl -L https://github.com/HocheggerLab/tcga-capstone-2026/archive/refs/heads/main.zip -o capstone.zip
unzip capstone.zip
cd tcga-capstone-2026-main
uv sync

uv sync installs exactly the packages the notebooks need. Open the folder in VS Code and select the .venv kernel. All the data is included, so it works offline.

Git version control for local development is explained in Lab 7. Prefer to browse first? See the repository on GitHub.

What you build

Every task is a function with a fixed output, and every function has a check cell underneath that runs it on a worked example — CDKN1A in breast cancer, assigned to nobody. Green ticks mean it works, before you run it on your own gene.

TaskFunctionPurpose
1aload_and_cleanClean the raw patient export into a tidy table.
1bdescribe_cohortSummarise the cohort as the paper’s “Table 1”.
2expression_landscapeRank your gene’s expression across cancer types.
3compare_groupsA Mann–Whitney U test — written once, used twice.
4plot_comparisonOne figure: normal, TP53 wild-type and TP53-mutant tumours.
5run_analysisCombine everything into one answer (marked per key).

Each function is marked on its own. Notebook 02 does not use Notebook 01 at all, so getting stuck on Task 1 never blocks the rest.

How it is marked

  • 20Function correctness — Each function meets its contract — marked per function.
  • 10Code quality — Single-purpose functions, no needless repetition.
  • 10Documentation — Docstrings, useful comments, names that say what they hold.
  • 20Interpretation — Table 1, the landscape, both comparisons, the figure, your answer.
  • 15GitHub repository — Repo link in Notebook 02, at least 5 meaningful commits, a README.
  • 25Extension (optional) — One option, done well — the route from 70 to 85.

A complete, well-written, well-interpreted core is first-class work and scores around 70. The extension is how you move from 70 to 85 — if you are struggling, do the core properly and skip it.

The extension — choose one

  • AThe tumours you called “wild-type” aren’t all wild-type — does TP53 deletion matter?Only for BRCA, LIHC, LUAD, PRAD and STAD.
  • BIs it really p53, or just gene dosage?
  • CDoes it hold in a different tissue?

Whichever you choose: the named function, the supplied checks passing, at least two useful checks of your own, and a short interpretation of what it changes about your Notebook 02 answer.

Your GitHub repository

  • →The link to your repo in the last cell of Notebook 02.
  • →At least 5 commits that capture meaningful steps, each with a brief message saying what changed and why.
  • →A README.md: the question, your gene and cancer type, how to run your notebooks, and your one-sentence answer.

In Colab, use File → Save a copy in GitHub at each milestone — every save is a commit. Working locally, git is covered in Lab 7.

About the data

Real, de-identified patient data from the open-access tiers of The Cancer Genome Atlas and the GTEx project. Treat it as you would your own medical record: never try to work out who a patient is. The full acknowledgements and citations are in 00_start_here.

Submit your notebooks through Canvas, where you will also find the deadline and your marks.