Input

Enter GEO IDs (one ID per line, maximum 10)



And/or upload your own data

Only tab-separated files (.txt and .tsv) are admitted Download a sample file (GPL570)

Analysis parameters

Meta-analysis method

Allowed missing values (%)


Adjusted P-value threshold


Group 1 name


Group 2 name


Functional analysis


Email



Loading...






Data source

GEO ID

Identifier of NCBI GEO dataset. It must begin with “GSE” followed by a numeric code (i.e. GSE10325). GSE file is a specific format from NCBI GEO which contains a gene expression matrix (Rows: genes or probe sets. Columns: samples) and all information about the experiment including annotation file and sample information. For more details, visit: http://www.ncbi.nlm.nih.gov/geo/info/datasets.html .

Your data

This web tool allows you to integrate your own data in the analysis. To upload the data, you must use the correct format (Figure 1). Your data must be saved in plain text format (.txt, or .tsv) where the first row contains “ID” in column one, followed by the sample counter in the next columns. The first value of the second row must be “NA” followed by a categorical vector (where value "0" is for group 1, "1" for group 2 and “X” if you want to exclude the sample from the analysis). The remaining rows contain the gene identifiers or the probe identifiers and, in the following columns, their expression values.

Figure 1: Data format to upload

Platforms admitted

Table 1 contains the admitted GEO platforms by ImaGEO. In addition, gene symbols can be used as gene identifiers for datasets submitted by yourself.


Table 1. Platforms admitted.

Meta-analysis overview

A meta-analysis is a statistical analysis that combines the results of multiple scientific studies. Gene expression meta-analysis combines different expression datasets from different sources. There are two major goals in meta-analysis studies: (1) Combine the same experimental condition across different studies to increase the sample size and the statistical power. (2) Compare different experimental conditions (i.e. different diseases) to discover common and different biomarkers.

Meta-analysis methods

Effect size combination

Effect size can mean different quantitative measurements to explain the strength of a phenomenon in different groups. In our case, effect size is the difference between two group means (i.e. case and control, case1 and case2, etc) divided by standard deviation, which are considered combinable and comparable across different studies. It is divided into two specific methods:

  1. Fixed Effect Model (FEM), where the estimated effect size in each study is assumed to come from an underlying true effect size plus measurement error.
  2. Random Effect Model (REM), where each study further contains a random effect that can incorporate unknown cross-study heterogeneities in the model (i.e. due to different platforms or different batches).

The selection of one or another method depends on the heterogeneity of the data. In the context of gene expression meta-analysis, a fixed-effects will identify the genes with strongest effect in the studies, while a random-effects model attempts to identify the genes with strongest average effect in a hypothetical population of studies. Although the latter is theoretically preferable, it might not be optimal if, for example, a source of heterogeneity is not expected to occur again in future data. In this context, a heterogeneity test can be useful to decide what is the most suitable method to use.

P-value combination

This method integrates the P-values from individual analysis into a combined P-value. There are different methods:

  1. Fisher's method: It is defined as summation of -log(P-value) across studies.
  2. Stouffer's method: Similar to Fisher's method but based on Z-scores, allowing incorporation of study weights. This method can be applied when all datasets have similar qualities (see Quality control section).
  3. maxP: This method is the most restrictive one. For each result, the maximum P-value is assigned across all studies.
  4. minP: The minimun of P-values across studies.
  5. AW: Adaptively-weighted method (Li and Tseng, 2011).
  6. SR: The sum of the ranks method.
  7. PR: The naive product of the ranks method. Similar to SR but computing the product instead of the sum.

These methods should give more significant results than effect size methods, but the confidence is smaller (greater number of potential false positives).

Combining p-values has an advantage for standardization of the associations from genomic studies to a common scale allowing to compare very heterogeneous datasets, for example datasets from different tissues.

Sample selection

Sample Selection tab shows four tables. The one at the top (Selected samples) shows the number of cases and controls selected for each dataset. Below this table there are the buttons to change the dataset and access the information of the samples in them. That information is shown in the second table at the left botton (Unassigned samples). In this table you have to select the samples and divide them according to the user criteria into the groups previously created (Cases/Controls in the example). The last two tables are specific to those groups. To select the samples you have to select them and click the right-oriented arrows related to the convenient table (Controls/Cases). Select them again in the group table and click the left-oriented to remove them for the analysis.

Figure 2: Sample selection

Meta-analysis results

  • ID: Gene symbol
  • Pval: P value is the probability for a given statistical model that, when the null hypothesis is true, the statistical summary (such as the sample mean difference between two compared groups) would be the same as or of greater magnitude than the actual observed results.
  • Fdr_pval: P-value adjusted for multiple testing by FDR method
  • FoldChange (only for P-value methods): fold changes are defined directly in terms of ratios. If the mean gene expression value in a set is A and for the other set is B, the fold change is defined as B/A.
  • Zval (only for Effect size methods): Z value is a test statistic for Z-tests that measure the difference between an observed statistic, the gene expression levels, and its hypothesized population parameter in units of the standard deviation.

Functional analysis results

  • GOBPID/GOMFID/GOCCID: Gene Ontology identifiers.
  • Pvalue: P-value for each term to be significantly enriched.
  • OddsRatio: the odds ratio for each category term tested
  • ExpCount: number of genes in the selected gene list to be found at each tested category term.
  • Count: number of genes from the gene set that are annotated at the term.
  • Size: number of genes annotated at the given GO term (where genes are restricted to the defined gene universe).
  • Term: descriptive name of the Gene Ontology identifiers.

Contact

If you have any doubt, question or suggestion, you can write us to bioinfo@genyo.es .



If you use ImaGEO, please include this reference:

Daniel Toro-Domínguez, Jordi Martorell-Marugán, Raúl López-Dominguez, Adrián García-Moreno, Víctor González-Rumayor, Marta E Alarcón-Riquelme, Pedro Carmona-Sáez (2018) ImaGEO: Integrative Gene Expression Meta-Analysis from GEO database. Bioinformatics. bty721, https://doi.org/10.1093/bioinformatics/bty721