---
title: "From GEO to downstream analysis"
author: "Sean Davis"
format:
  html:
    toc: true
vignette: >
  %\VignetteIndexEntry{From GEO to downstream analysis}
  %\VignetteEngine{quarto::html}
  %\VignetteEncoding{UTF-8}
---

GEOquery's job ends once your data is a Bioconductor object. This article is a map
of *where to go next* — how the object you got from `getGEO()` connects to the
rest of the Bioconductor ecosystem — rather than a tutorial for any one method.
The goal is to save you the "I have a `SummarizedExperiment`, now what?" moment.

## Know your object

What you do next depends on what GEOquery returned (see
[Understanding GEO data formats](geo-data-formats.html)):

| You have | Typical source | Downstream entry point |
|----------|----------------|------------------------|
| `SummarizedExperiment` (list) | GSE Series Matrix (default), RNA-seq counts | limma / DESeq2 / edgeR |
| `ExpressionSet` (list) | GSE Series Matrix with `returnType = "ExpressionSet"` | limma |
| `SingleCellExperiment` | single-cell readers | scater / scran / OSCA |
| `GSE`/`GSM`/`GPL`/`GDS` S4 | SOFT parsing | accessors, then convert |

## Microarray: limma

Microarray Series Matrix data arrives already processed (typically
log-transformed, normalized intensities). The standard path is a linear model
with [limma](https://bioconductor.org/packages/limma), which accepts a matrix:


``` r
library(limma)
library(SummarizedExperiment)
se <- getGEO("GSE2553")[[1]]                 # SummarizedExperiment (default)

# `group` here stands in for a real grouping column you derive from the sample
# metadata -- see the note below; it will not exist verbatim in colData(se).
design <- model.matrix(~ group, data = colData(se))
fit <- eBayes(lmFit(assay(se), design))
topTable(fit, coef = 2)
```

The hardest part is usually not the model but extracting clean grouping variables
from `colData(se)` — GEO sample metadata is free text, so expect to parse
`characteristics_ch1` fields into the factor you actually model.

## RNA-seq counts: DESeq2 / edgeR / limma-voom

For NCBI-computed RNA-seq counts (see the [RNA-seq article](rnaseq.html)), use a
count-based model. The `SummarizedExperiment` GEOquery returns plugs directly into
[DESeq2](https://bioconductor.org/packages/DESeq2):


``` r
library(DESeq2)
se <- getRNASeqData("GSE164073")
# `condition` must be a real column you added to colData(se) first.
dds <- DESeqDataSet(se, design = ~ condition)
dds <- DESeq(dds)
results(dds)
```

[edgeR](https://bioconductor.org/packages/edgeR) and
[limma-voom](https://bioconductor.org/packages/limma) are equally good choices and
consume the same count matrix.

## Single cell: the OSCA stack

A `SingleCellExperiment` from the [single-cell readers](single-cell.html) is the
entry point to Bioconductor's single-cell ecosystem —
[scater](https://bioconductor.org/packages/scater),
[scran](https://bioconductor.org/packages/scran), and the
[OSCA book](https://bioconductor.org/books/release/OSCA/) for the full
quality-control → normalization → clustering → annotation arc.

## Converting and annotating

Two recurring needs:

- **Modernize a legacy result.** `getGEO()` returns `SummarizedExperiment` by
  default now, but if you have an older `ExpressionSet` (or asked for one with
  `returnType = "ExpressionSet"`), convert it without re-downloading:

  
  ``` r
  se <- as_SummarizedExperiment(getGEO("GSE2553", returnType = "ExpressionSet")[[1]])
  ```

- **Re-annotate features.** GEO platform annotation can be dated. For mapping
  probe/gene identifiers, reach for the Bioconductor annotation infrastructure —
  [AnnotationDbi](https://bioconductor.org/packages/AnnotationDbi), organism
  packages such as `org.Hs.eg.db`, and
  [biomaRt](https://bioconductor.org/packages/biomaRt).

## Reproducibility

GEO records can change and large downloads are slow, so cache deliberately and
record versions. The [Finding and downloading data](finding-and-downloading-data.html)
article covers persistent caching (`destdir=` and the `BiocFileCache` layer);
alongside that, capture `sessionInfo()` with your results so the GEOquery and
dependency versions are pinned.

## Where to go next

- [Finding and downloading data](finding-and-downloading-data.html) — caching and
  reproducible retrieval.
- [RNA-seq](rnaseq.html) and [Single-cell](single-cell.html) — the data-type
  paths that feed these workflows.