About Gemma

Gemma is a web site, database and a set of tools for the meta-analysis, re-use and sharing of genomics data, currently primarily targeted at the analysis of gene expression profiles. Gemma contains data from thousands of public studies, referencing thousands of published papers. Every dataset in Gemma has passed a rigorous curation process that re-annotates the expression platform at the sequence level, which allows for more consistent cross-platform comparisons and meta-analyses.

For detailed information on the curation process, read this page or the latest publication.

Package cheat sheet

Installation instructions

if (!require("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

BiocManager::install("gemma.R")

Searching for datasets of interest in Gemma

Using the get_datasets function, datasets fitting various criteria can be accessed.

# accessing all mouse and human datasets
get_datasets(taxa = c('mouse','human')) %>% 
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

# accessing human datasets with the word "bipolar"
get_datasets(query = 'bipolar',taxa = 'human') %>% 
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

# access human datasets that were annotated with the ontology term for the
# bipolar disorder
# use search_annotations function to search for available annotation terms
get_datasets(taxa ='human', 
             uris = 'http://purl.obolibrary.org/obo/MONDO_0004985') %>%
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

get_dataset function also includes a filter parameter that allows filtering for datasets with specific properties in a more structured manner. A list of the available properties can be accessed using filter_properties

filter_properties()$dataset %>% head %>% gemma_kable()

These properties can be used together to fine tune your results

# access human datasets that has bipolar disorder as an experimental factor
get_datasets(taxa = 'human',
             filter = "experimentalDesign.experimentalFactors.factorValues.characteristics.valueUri = http://purl.obolibrary.org/obo/MONDO_0004985")  %>%
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

# all datasets with more than 4 samples annotated for any disease
get_datasets(filter = 'bioAssayCount > 4 and allCharacteristics.category = disease') %>%
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

# all datasets with ontology terms for Alzheimer's disease and Parkinson's disease
# this is equivalent to using the uris parameter
get_datasets(filter = 'allCharacteristics.valueUri in (http://purl.obolibrary.org/obo/MONDO_0004975,http://purl.obolibrary.org/obo/MONDO_0005180
)')  %>%
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

Note that a single call of these functions will only return 20 results by default and a 100 results maximum, controlled by the limit argument. In order to get all available results, use get_all_pages function on the output of the function

get_datasets(taxa = 'human') %>% 
    get_all_pages() %>% 
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

See larger queries section for more details. To keep this vignette simpler we will keep using the first 20 results returned by default in examples below.

Dataset information provided by get_datasets also includes some quality information that can be used to determine the suitability of any given experiment. For instance experiment.batchEffect column will be set to -1 if Gemma’s preprocessing has detected batch effects that were unable to be resolved by batch correction. More information about these and other fields can be found at the function documentation.

get_datasets(taxa = 'human', filter = 'bioAssayCount > 4') %>% 
     filter(experiment.batchEffect !=-1) %>% 
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

Gemma uses multiple ontologies when annotating datasets and using the term URIs instead of free text to search can lead to more specific results. search_annotations function allows searching for annotation terms that might be relevant to your query.

search_annotations('bipolar') %>% 
    head %>% gemma_kable()

Downloading expression data

Upon identifying datasets of interest, more information about specific ones can be requested. In this example we will be using GSE46416 which includes samples taken from healthy donors along with manic/euthymic phase bipolar disorder patients.

The data associated with specific experiments can be accessed by using get_datasets_by_ids

get_datasets_by_ids("GSE46416") %>%
    select(experiment.shortName, experiment.name, 
           experiment.description,taxon.name) %>%
    head %>% gemma_kable

To access the expression data in a convenient form, you can use get_dataset_object. It is a high-level wrapper that combines various endpoint calls to return lists of annotated SummarizedExperiment or ExpressionSet objects that are compatible with other Bioconductor packages or a tidyverse-friendly long form tibble for downstream analyses. These include the expression matrix along with the experimental design, and ensure the sample names match between both when transforming/subsetting data.

dat <- get_dataset_object("GSE46416",
                          type = 'se') # SummarizedExperiment is the default output type

Note that the tidy format is less memory efficient but allows easy visualization and exploration with ggplot2 and the rest of the tidyverse.

To show how subsetting works, we’ll keep the “manic phase” data and the reference_subject_roles, which refers to the control samples in Gemma datasets.

# Check the levels of the disease factor
dat[[1]]$disease %>% unique()

[1] "bipolar disorder has_modifier euthymic phase"
[2] "reference subject role"                      
[3] "bipolar disorder has_modifier manic phase"   

# Subset patients during manic phase and controls
manic <- dat[[1]][, dat[[1]]$disease == "bipolar disorder has_modifier manic phase" | 
        dat[[1]]$disease == "reference subject role"]
manic

class: SummarizedExperiment 
dim: 18758 21 
metadata(8): title abstract ... gemmaSuitabilityScore taxon
assays(1): counts
rownames(18758): 2315430 2315554 ... 7385683 7385696
rowData names(4): Probe GeneSymbol GeneName NCBIid
colnames(21): Control, 12 Control, 1_DE50 ... Bipolar disorder patient
  manic phase, 21 Bipolar disorder patient manic phase, 16
colData names(3): factorValues block disease

Let’s take a look at sample to sample correlation in our subset.

# Get Expression matrix
manicExpr <- assay(manic, "counts")


manicExpr %>% 
    cor %>% 
    pheatmap(col =viridis(10),border_color = NA,angle_col = 45,fontsize = 7)

Sample to sample correlations of bipolar patients during manic phase and controls.

You can also use get_dataset_processed_expression to only get the expression matrix, get_dataset_samples to get the metadata information. The output of this function includes some additional details about a sample such as the original accession ID or whether or not it was determined to be an outlier but it can be simplified to match the design table included in the output of get_dataset_object by using make_design on the output.

get_dataset_samples('GSE46416') %>% make_design('text') %>% select(-factorValues) %>%  head %>%
    gemma_kable()

Platform Annotations

Expression data in Gemma comes with annotations for the gene each expression profile corresponds to. Using the get_platform_annotations function, these annotations can be retrieved independently of the expression data, along with additional annotations such as Gene Ontology terms.

Examples:

head(get_platform_annotations('GPL96') %>% select(-GOTerms))

   ElementName   GeneSymbols                                     GeneNames
        <char>        <char>                                        <char>
1: 211750_x_at TUBA1C|TUBA1A             tubulin alpha 1c|tubulin alpha 1a
2:   216678_at                                                            
3:   216345_at        ZSWIM8            zinc finger SWIM-type containing 8
4:   207273_at                                                            
5: 216025_x_at        CYP2C9 cytochrome P450 family 2 subfamily C member 9
6: 218191_s_at        LMBRD1                     LMBR1 domain containing 1
        GemmaIDs    NCBIids
          <char>     <char>
1: 360802|172797 84790|7846
2:                         
3:        235733      23053
4:                         
5:         32964       1559
6:        303717      55788

head(get_platform_annotations('Generic_human_ncbiIds') %>% select(-GOTerms))

   ElementName  GeneSymbols
         <int>       <char>
1:       55236         UBA6
2:       79664         ICE2
3:   100126270     FMR1-AS1
4:   105373684    LINC01818
5:   124900245 LOC124900245
6:   102725051 LOC102725051
                                               GeneNames GemmaIDs   NCBIids
                                                  <char>    <int>     <int>
1:           ubiquitin like modifier activating enzyme 6   295849     55236
2: interactor of little elongation complex ELL subunit 2   336840     79664
3:                                  FMR1 antisense RNA 1  3157248 100126270
4:           long intergenic non-protein coding RNA 1818  9235895 105373684
5:                           small nucleolar RNA SNORA40 10578422 124900245
6:                          uncharacterized LOC102725051  9008981 102725051

If you are interested in a particular gene, you can see which platforms include it using get_gene_probes. Note that functions to search gene work best with unambigious identifiers rather than symbols.

# lists genes in gemma matching the symbol or identifier
get_genes('Eno2') %>% gemma_kable()

# ncbi id for human ENO2
probes <- get_gene_probes(2026)

# remove the description for brevity of output
head(probes[,.SD, .SDcols = !colnames(probes) %in% c('mapping.Description','platform.Description')]) %>%
    gemma_kable()

Differential expression analyses

Gemma contains precomputed differential expression analyses for most of its datasets. Analyses can involve more than one factor, such as “sex” as well as “disease”. Some datasets contain more than one analysis to account for different factors and their interactions. The results are stored as resultSets, each corresponding to one factor (or their interaction). You can access them using get_differential_expression_values. From here on, we can explore and visualize the data to find the most differentially-expressed genes

Note that get_differential_expression_values can return multiple differentials per study if a study has multiple factors to contrast. Since GSE46416 only has one extracting the first element of the returned list is all we need.

dif_exp <- get_differential_expression_values('GSE46416')
dif_exp[[1]] %>% head %>% gemma_kable()

By default the columns names of the output correspond to contrast IDs. To see what conditions these IDs correspond to we can either use get_dataset_differential_expression_analyses to get the metadata about differentials of a given dataset, or set readableContrasts argument of get_differential_expression_values to TRUE. The former approach is usually better for a large scale systematic analysis while the latter is easier to read in an interactive session.

get_dataset_differential_expression_analyses function returns structured metadata about the differentials.

contrasts <- get_dataset_differential_expression_analyses('GSE46416')
contrasts %>% gemma_kable()

contrast.ID column corresponds to the column names in the output of get_differential_expression_values while result.ID corresponds to the name of the differential in the output object. Using them together will let one to access differentially expressed gene counts for each condition contrast

# using result.ID and contrast.ID of the output above, we can access specific
# results. Note that one study may have multiple contrast objects
seq_len(nrow(contrasts)) %>% sapply(function(i){
    result_set = dif_exp[[as.character(contrasts[i,]$result.ID)]]
    p_values = result_set[[glue::glue("contrast_{contrasts[i,]$contrast.ID}_pvalue")]]
    
    # multiple testing correction
    sum(p.adjust(p_values,method = 'BH') < 0.05)
}) -> dif_exp_genes

contrasts <- data.table(result.ID = contrasts$result.ID,
                        contrast.id = contrasts$contrast.ID,
                        baseline.factorValue = contrasts$baseline.factors,
                        experimental.factorValue = contrasts$experimental.factors,
                        n_diff = dif_exp_genes)

contrasts %>% gemma_kable()

contrasts$baseline.factors

NULL

contrasts$experimental.factors

NULL

Alternatively we, since we are only looking at one dataset and one contrast manually, we can simply use readableContrasts.

de <- get_differential_expression_values("GSE46416",readableContrasts = TRUE)[[1]]
de %>% head %>% gemma_kable

# Classify probes for plotting
de$diffexpr <- "No"
de$diffexpr[de$`contrast_bipolar disorder has_modifier manic phase_logFoldChange` > 1.0 & 
        de$`contrast_bipolar disorder has_modifier manic phase_pvalue` < 0.05] <- "Up"
de$diffexpr[de$`contrast_bipolar disorder has_modifier manic phase_logFoldChange` < -1.0 & 
        de$`contrast_bipolar disorder has_modifier manic phase_pvalue` < 0.05] <- "Down"

# Upregulated probes
filter(de, diffexpr == "Up") %>%
    arrange(`contrast_bipolar disorder has_modifier manic phase_pvalue`) %>%
    select(Probe, GeneSymbol, `contrast_bipolar disorder has_modifier manic phase_pvalue`, 
        `contrast_bipolar disorder has_modifier manic phase_logFoldChange`) %>%
    head(10) %>% gemma_kable()

# Downregulated probes
filter(de, diffexpr == "Down") %>%
    arrange(`contrast_bipolar disorder has_modifier manic phase_pvalue`) %>%
    select(Probe, GeneSymbol, `contrast_bipolar disorder has_modifier manic phase_pvalue`, 
        `contrast_bipolar disorder has_modifier manic phase_logFoldChange`) %>%
    head(10) %>% gemma_kable()

# Add gene symbols as labels to DE genes
de$delabel <- ""
de$delabel[de$diffexpr != "No"] <- de$GeneSymbol[de$diffexpr != "No"]

# Volcano plot for bipolar patients vs controls
ggplot(
    data = de,
    aes(
        x = `contrast_bipolar disorder has_modifier manic phase_logFoldChange`,
        y = -log10(`contrast_bipolar disorder has_modifier manic phase_pvalue`),
        color = diffexpr,
        label = delabel
    )
) +
    geom_point() +
    geom_hline(yintercept = -log10(0.05), col = "gray45", linetype = "dashed") +
    geom_vline(xintercept = c(-1.0, 1.0), col = "gray45", linetype = "dashed") +
    labs(x = "log2(FoldChange)", y = "-log10(p-value)") +
    scale_color_manual(values = c("blue", "black", "red")) +
    geom_text_repel(show.legend = FALSE) +
    theme_minimal()

Differentially-expressed genes in bipolar patients during manic phase versus controls.

Larger queries

To query large amounts of data, the API has a pagination system which uses the limit and offset parameters. To avoid overloading the server, calls are limited to a maximum of 100 entries, so the offset allows you to get the next batch of entries in the next call(s).

To simplify the process of accessing all available data, gemma.R includes the get_all_pages function which can use the output from one page to make all the follow up requests

get_platforms_by_ids() %>% 
    get_all_pages() %>% head %>% gemma_kable()

Alternative way to access all pages is to do so manually. To see how many available results are there, you can look at the attributes of the output objects where additional information from the API response is appended.

platform_count = attributes(get_platforms_by_ids(limit = 1))$totalElements
print(platform_count)

[1] 634

After which you can use offset to access all available platforms.

lapply(seq(0,platform_count,100), function(offset){
    get_platforms_by_ids(limit = 100, offset = offset) %>%
        select(platform.ID, platform.shortName, taxon.name)
}) %>% do.call(rbind,.) %>% 
    head %>% gemma_kable()

Many endpoints only support a single identifier:

get_dataset_annotations(c("GSE35974", "GSE46416"))

Error: Please specify one valid identifier for dataset.

In these cases, you will have to loop over all the identifiers you wish to query and send separate requests.

lapply(c("GSE35974", "GSE12649"), function(dataset) {
    get_dataset_annotations(dataset) %>% 
        mutate(experiment.shortName = dataset) %>%
        select(experiment.shortName, class.name, term.name)
}) %>% do.call(rbind,.) %>% gemma_kable()

Output options

get_gene_locations("DYRK1A") %>% gemma_kable()

get_gene_locations("DYRK1A", raw = TRUE) %>% jsonedit()

# use memoisation by default using the default cache
gemma_memoised(TRUE)

# set an altnernate cache path
gemma_memoised(TRUE,"path/to/cache_directory")

# cache in memory of the R session
# this cache will not be preserved between sessions
gemma_memoised(TRUE,"cache_in_memory")

options(gemma.memoised = TRUE) # always refer to cache. this is redundant with gemma_memoised function
options(gemma.overwrite = TRUE) # always overwrite when saving files
options(gemma.raw = TRUE) # always receive results as-is from Gemma

Session info

sessionInfo()

R version 4.4.1 (2024-06-14)
Platform: x86_64-pc-linux-gnu
Running under: Ubuntu 24.04.1 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

time zone: Etc/UTC
tzcode source: system (glibc)

attached base packages:
[1] stats4    stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
 [1] listviewer_4.0.0            viridis_0.6.5              
 [3] viridisLite_0.4.2           pheatmap_1.0.12            
 [5] SummarizedExperiment_1.35.5 Biobase_2.67.0             
 [7] GenomicRanges_1.57.2        GenomeInfoDb_1.41.2        
 [9] IRanges_2.39.2              S4Vectors_0.43.2           
[11] BiocGenerics_0.53.0         MatrixGenerics_1.17.1      
[13] matrixStats_1.4.1           ggrepel_0.9.6              
[15] ggplot2_3.5.1               dplyr_1.1.4                
[17] data.table_1.16.2           gemma.R_3.3.0              
[19] BiocStyle_2.35.0           

loaded via a namespace (and not attached):
 [1] tidyselect_1.2.1        farver_2.1.2            fastmap_1.2.0          
 [4] digest_0.6.37           timechange_0.3.0        lifecycle_1.0.4        
 [7] magrittr_2.0.3          compiler_4.4.1          rlang_1.1.4            
[10] sass_0.4.9              tools_4.4.1             utf8_1.2.4             
[13] yaml_2.3.10             knitr_1.48              labeling_0.4.3         
[16] S4Arrays_1.5.11         htmlwidgets_1.6.4       bit_4.5.0              
[19] curl_5.2.3              DelayedArray_0.33.1     xml2_1.3.6             
[22] RColorBrewer_1.1-3      abind_1.4-8             withr_3.0.2            
[25] purrr_1.0.2             sys_3.4.3               grid_4.4.1             
[28] fansi_1.0.6             colorspace_2.1-1        scales_1.3.0           
[31] cli_3.6.3               rmarkdown_2.28          crayon_1.5.3           
[34] generics_0.1.3          rstudioapi_0.17.1       httr_1.4.7             
[37] cachem_1.1.0            stringr_1.5.1           zlibbioc_1.51.2        
[40] assertthat_0.2.1        BiocManager_1.30.25     XVector_0.45.0         
[43] vctrs_0.6.5             Matrix_1.7-1            jsonlite_1.8.9         
[46] bit64_4.5.2             systemfonts_1.1.0       maketools_1.3.1        
[49] jquerylib_0.1.4         glue_1.8.0              lubridate_1.9.3        
[52] stringi_1.8.4           gtable_0.3.6            UCSC.utils_1.1.0       
[55] munsell_0.5.1           tibble_3.2.1            pillar_1.9.0           
[58] rappdirs_0.3.3          htmltools_0.5.8.1       GenomeInfoDbData_1.2.13
[61] R6_2.5.1                evaluate_1.0.1          kableExtra_1.4.0       
[64] lattice_0.22-6          highr_0.11              memoise_2.0.1          
[67] bslib_0.8.0             Rcpp_1.0.13             svglite_2.1.3          
[70] gridExtra_2.3           SparseArray_1.5.45      xfun_0.48              
[73] buildtools_1.0.0        pkgconfig_2.0.3