GEO Database Guide: Finding Gene Expression Studies You Can Reanalyze

The Gene Expression Omnibus (GEO) at NCBI is the world’s largest public repository for gene expression data, hosting over four million samples from microarray and high throughput sequencing experiments. This guide shows you how to search GEO for studies suitable for reanalysis, inspect sample metadata for compatibility, download the correct files, and avoid the common pitfalls that lead to mismatched comparisons. You should read this if you are a bioinformatician looking for validation data, a bench scientist wanting to reuse published results, or a student learning how to work with public transcriptomics datasets. For an introduction to how GEO data is structured and curated, refer to the NCBI Bookshelf section on the Gene Expression Omnibus.

At a Glance

Use the official EMBL EBI Training resources for complementary tutorials on searching and downloading functional genomics data.

Decision Criteria for Selecting a Study for Reanalysis

Not every GEO record is suitable for reuse. Apply these criteria before investing time in a dataset.

1. Raw data must be accessible. For microarrays, look for CEL, TXT, or GPR files. For sequencing, the raw reads should be deposited in the Sequence Read Archive (SRA) and linked from the GEO entry. The NCBI Sequence Read Archive provides a direct path to download raw sequencing data when the GEO record indicates that data exists in SRA.

2. Sample metadata need to be interpretable. Each GSM entry should include a clear description of the biological condition (e.g., “tumor stage II”, “control untreated”) and the organism, tissue, and any treatment details. Avoid studies that use cryptic abbreviations or lack a “characteristics” field.

3. The platform is still supported. GEO platform records (GPL) contain annotation information. If the platform is discontinued and no current annotation file exists, mapping probe IDs to gene symbols will be difficult. Check the GPL page for an updated annotation file or a note about replacement platforms.

4. Sufficient sample size is present. For differential expression, aim for at least three biological replicates per condition. Smaller studies may produce underpowered results that do not replicate.

5. The study design matches your question. A study labeled “expression profiling of breast cancer” may include only tumor samples with no normal tissue comparator. Verify the full experimental design on the GSE summary page before proceeding.

Practical Workflow: From Search to Download

Follow this step by step process to identify and acquire usable GEO data.

Step 1: Search GEO Using Entrez

Go to the GEO homepage at ncbi.nlm.nih.gov/geo and use the “Search GEO” box. You can combine terms with Boolean operators. For example, “(prostate cancer) AND (homo sapiens) AND (expression profiling by high throughput sequencing)” returns RNA seq studies. Limit results to “Series” (GSE) to see study level records. The Galaxy Training Network offers a module on searching biological databases that complements this step.

Step 2: Inspect the GSE Summary Page

Click on a GSE accession. The summary page includes the title, overall design, contributor list, and submission date. Scroll to the “Samples” table at the bottom. This table lists all GSM IDs, their titles, and the raw data download links. Verify that the sample titles and descriptions match the experimental groups you need.

Step 3: Examine Individual GSM Records

Click a few representative GSM numbers. The GSM record contains the platform (GPL), the organism, the processed data value table, and the “characteristics” field (metadata). Look for complete and consistent metadata across all samples. If some samples are missing treatment information or have conflicting descriptions, the study may be problematic for meta analysis. A system for systematic quality assessment is described in the EMERALD project paper Contributions of the EMERALD project to assessing and improving microarray data quality.

Step 4: Decide on Data Level

You need to decide whether to work with processed data (e.g., series matrix files) or raw data. Processed data are ready for analysis but are often normalized by methods unknown to you. Raw data allow you to apply your own normalization pipeline. For sequencing, raw FASTQ files from SRA are the gold standard. For microarrays, you may prefer raw CEL files.

Step 5: Download Data

For microarrays, use the “Supplementary file” links on the GSE page. A batch download can be performed using the FTP link (often ftp.ncbi.nlm.nih.gov/geo/...). For sequencing, the SRA Run Selector tool (linked from the GSE page) lets you download SRA files which can then be converted to FASTQ using the SRA Toolkit. Alternatively, the Bioconductor package GEOquery automates the download of series matrices and raw data in R. The BART tool also simplifies access BART: bioinformatics array research tool.

Step 6: Prepare Data for Reanalysis

After downloading, verify the file integrity. For microarrays, read the CEL files with a suitable package (e.g., affy in R). For RNA seq, align reads to the appropriate reference genome. Always document the version of platform annotation you use. The EMAAS system provides a grid based approach for managing and analyzing such data EMAAS: an extensible grid based rich internet application for microarray data analysis and management.

Common Mistakes That Ruin Reanalysis

Mistake 1: Assuming all samples in a GSE are comparable. Different GSM records within the same GSE may have been processed on different days or with different reagents. Check the “Protocol” fields for any notes about batch.

Mistake 2: Ignoring platform differences. Two studies on the same GPL can have different versions of the annotation file. Always use the annotation from the same date as the study submission if available.

Mistake 3: Overlooking normalization. Series matrix files are usually normalized, but the normalization method (RMA, MAS5, quantile, etc.) is not always stated. Reanalyzing normalized data without knowing the method can lead to artefacts.

Mistake 4: Using the wrong data type for comparison. Do not mix one channel and two channel microarray data in the same analysis without a proper transformation.

Mistake 5: Forgetting to check for missing metadata. Some studies do not include the control group information in the sample title. You may need to read the full publication to map sample IDs to conditions.

Limits and Uncertainty in GEO Reanalysis

GEO is a heterogeneous repository with no standardized quality control across submissions. A study may appear complete but the raw data could be corrupted or the metadata ambiguous. Furthermore, platforms become obsolete, and annotation may be incomplete. Studies that rely on older microarray platforms (e.g., Affymetrix Human Genome U95) require careful handling because many probes are not mapped to current gene symbols. Sequencing data from SRA often lacks consistent quality metrics, and you must run your own QC before analysis.

Another limitation is the presence of duplicate samples or mislabeled files. The same sample may appear under multiple GSM IDs. Cross referencing with the associated publication is essential. The context sensitive approach described in Context sensitive autoassociative memories as expert systems in medical diagnosis illustrates the importance of contextual metadata for reliable interpretation.

Finally, reanalysis of public data always carries the risk of hidden batch effects. After combining multiple studies, always test for batch effects using methods such as PCA or ComBat. For spatial analysis of expression data, approaches from other fields may be adapted Spatial analysis for psychologists: How to use individual level data for research at the geographically aggregated level.

Frequently Asked Questions

1. Can I reanalyze any GEO dataset even if it lacks raw data? Yes, you can use processed series matrix files for many analyses. However, you will be limited to the normalization method chosen by the original submitter. For full flexibility, choose studies that include raw data (CEL or FASTQ).

2. How do I find raw sequencing data within GEO? Look for a “SRA” link on the GSE page. That link directs you to the SRA project. Alternatively, the GSM records for sequencing studies often include a “sra” or “fastq” link. Use the SRA Toolkit to download.

3. What should I do if the sample metadata is incomplete? First, read the associated publication (usually listed on the GSE page). If the paper is not available, you may need to contact the authors. Some studies have metadata only in the “characteristics” field as free text, which can be parsed manually.

4. Can I combine two GEO studies that used the same platform? Yes, but only if you can confirm that the platform version is identical and that the data were generated under similar conditions. You must also test and correct for batch effects. Combining datasets from different labs increases the risk of technical artifacts.

References and Further Reading

1. NCBI Bookshelf: GEO User Guide https://www.ncbi.nlm.nih.gov/books/
2. EMBL EBI Training: Functional genomics data resources https://www.ebi.ac.uk/training/
3. Galaxy Training Network: Searching biological databases https://training.galaxyproject.org/
4. Bioconductor: GEOquery and analysis packages https://bioconductor.org/
5. NCBI Sequence Read Archive https://www.ncbi.nlm.nih.gov/sra
6. Contributions of the EMERALD project to assessing and improving microarray data quality https://pubmed.ncbi.nlm.nih.gov/21231919/
7. BART: bioinformatics array research tool https://pubmed.ncbi.nlm.nih.gov/30089462/
8. EMAAS: an extensible grid based rich internet application for microarray data analysis and management https://pubmed.ncbi.nlm.nih.gov/19032776/
9. CellChat for systematic analysis of cell cell communication from single cell transcriptomics https://pubmed.ncbi.nlm.nih.gov/39289562/
10. Spatial analysis for psychologists: How to use individual level data for research at the geographically aggregated level https://pubmed.ncbi.nlm.nih.gov/35653725/

Related Articles

• RNA Sequencing Analysis: From FASTQ Files to Biological Questions
• RNA-seq Quality Control: What to Check Before Differential Expression
• How to Plan a Bulk RNA-seq Differential Expression Study
• Single-Cell RNA-seq Workflow: A Practical Analysis Roadmap
• Single-Cell RNA-seq Quality Control: Cells, Genes, and Mitochondrial Reads