UMass Chan Bioinformatics CoreBioinformatics Core
GitHub

Bootcamp / Session 2 / 2.1

Session 2.1 · Ming “Tommy” Tang, AstraZeneca

Basic statistics for biologists.

What a p-value actually means, why testing twenty thousand genes at once breaks it, and how false discovery rate, adjusted p-values, and q-values put it back together. The statistics you need to judge a gene list, your own or one an AI hands you.

Open the slides Fri Sep 25, 1:00 to 4:00 pm · Slides are still being written and will change before the session

Recordings

The afternoon opened with a short welcome, then Tommy's statistics talk.

Welcome

Alper Kucukural · 3 minutes · Download (14 MB)

Basic statistics for biologists

Tommy Tang · 36 minutes · Download (80 MB)

What the talk covers

  1. Start from the null hypothesis. Every p-value begins by assuming one, so you cannot read the number without stating it.
  2. What a p-value is. The probability of seeing a difference this large if the null were true, and what that does not tell you.
  3. P-value distributions. What the histogram of all your p-values says about the experiment before you look at any single gene.
  4. The multiple comparison problem. Thousands of genes tested at once, and what happens to your false positives.
  5. Controlling the false discovery rate. The Benjamini and Hochberg method, and what it actually promises.
  6. Adjusted p-values and q-values. What p.adjust() recalculates, and how a q-value differs.
  7. Large samples always look significant. With enough points, a meaningless difference still clears p < 0.05, which is why effect size matters.
  8. Single-cell data. Where thousands of cells make that problem worse.
  9. Small samples mislead. Why a boxplot over a handful of samples can show you a pattern that is not there.

Why this comes right after Session 1

Session 1 ended on a rule: verify what AI produces. This is the part of the toolkit that lets you do it on a result.

An AI assistant will confidently run the wrong test, or hand you a gene list with uncorrected p-values, and the output looks exactly as polished either way.

Knowing what a p-value promises is how you catch that. Statistics is the detector.

Materials

SlidesBasic statistics for biologistsTommy's deck for this part. Work in progress.docs.google.com ReadingP-values, multiple comparisons, FDR, and q-valuesTommy's written version of this talk, with the worked arithmetic.divingintogeneticsandgenomics.com ReadingHow to interpret a p-value histogramDavid Robinson. What the shape of the distribution is telling you.varianceexplained.org ReadingCommon statistical tests are linear modelsOne framework behind the tests you were taught as separate recipes.lindeloev.github.io ReadingIntroduction to linear mixed modelsFor designs with repeated measures or nested samples.gkhajduk.github.io ReadingPoints of SignificanceThe Nature Methods statistics column, collected.nature.com

After the session

Two minutes of feedback, so the next session is pitched better. It also records your attendance, so fill it in whether you were in the room, on Zoom, or catching up on the recording.

Session 2 feedback