GOfuncR: Gene Ontology Enrichment Using FUNC
May 23, 2021
Contents
- 1 Overview
- 2 Examples of GO-enrichment analyses for human genes
- 3 Enrichment analyses with different annotations or ontologies
- 4 Additional functionalities
- 5 Schematics
- 6 Session Info
- 7 References
1 Overview
GOfuncR performs a gene ontology enrichment analysis based on the ontology enrichment software FUNC [1,2].
It provides the standard candidate vs. background enrichment analysis using the hypergeometric test, as well as three additional tests: (i) the Wilcoxon rank-sum test that is used when genes are ranked, (ii) a binomial test that can be used when genes are associated with two counts, e.g. amino acid changes since a common ancestor in two different species, and (iii) a 2x2 contingency table test that is used in cases when genes are associated with four counts, e.g. non-synonymous or synonymous variants that are fixed between or variable within species.
To correct for multiple testing and interdependency of the tests, family-wise error rates (FWER) are computed based on random permutations of the gene-associated variables (see Schematic 1 below).
GOfuncR also provides tools for exploring the ontology graph and the annotations, and options to take gene-length or spatial clustering of genes into account during testing.
GO-annotations and gene-coordinates are obtained from OrganismDb packages (Homo.sapiens by default) or OrgDb and TxDb packages.
The gene ontology graph (obtained from geneontology, release date 01-May-2021), is integrated in the package.
It is also possible to provide custom gene coordinates, annotations and ontologies.
input data and test selection
1.1 Functions included in GOfuncR
| function | description |
|---|---|
| go_enrich | core function for performing enrichment analyses given a candidate gene set |
| plot_anno_scores | plots distribution of scores of genes annotated to GO-categories |
| get_parent_nodes | returns all parent-nodes of input GO-categories |
| get_child_nodes | returns all child-nodes of input GO-categories |
| get_names | returns the full names of input GO-categories |
| get_ids | returns all GO-categories that contain the input string |
| get_anno_genes | returns genes that are annotated to input GO-categories |
| get_anno_categories | returns GO-categories that input genes are annotated to |
| refine | restrict results to most specific GO-categories |
1.2 Core function go_enrich
The function go_enrich performs all enrichment analyses given input genes and has the following parameters:
| parameter | default | description |
|---|---|---|
genes |
- | a dataframe with gene-symbols or genomic regions and gene-associated variables |
test |
‘hyper’ | statistical test to use (‘hyper’, ‘wilcoxon’, ‘binomial’ or ‘contingency’) |
n_randsets |
1000 | number of randomsets for computing the family-wise error rate |
organismDb |
‘Homo.sapiens’ | OrganismDb package for GO-annotations and gene coordinates |
gene_len |
FALSE | correct for gene length (only for test='hyper') |
regions |
FALSE | chromosomal regions as input instead of independent genes (only for test='hyper') |
circ_chrom |
FALSE | use background on circularized chromosome (only for test='hyper' and regions=TRUE) |
silent |
FALSE | suppress output to screen |
domains |
NULL | optional vector of GO-domains (if NULL all 3 domains are analyzed) |
orgDb |
NULL | optional OrgDb package for GO-annotations (overrides organismDb) |
txDb |
NULL | optional TxDb package for gene-coordinates (overrides organismDb) |
annotations |
NULL | optional dataframe with GO-annotations (overrides organismDb and orgDb) |
gene_coords |
NULL | optional dataframe with gene-coordinates (overrides organismDb and txDb) |
godir |
NULL | optional directory with ontology graph tables to use instead of the integrated GO-graph |
2 Examples of GO-enrichment analyses for human genes
2.1 Install annotation package
GOfuncR uses external packages to obtain the GO-annotations and gene-coordinates.
In the examples we will use the default Homo.sapiens package.
See below for examples how to use other packages or how to provide custom annotations.
## install annotation package 'Homo.sapiens' from bioconductor
if (!requireNamespace("BiocManager", quietly=TRUE))
install.packages("BiocManager")
BiocManager::install('Homo.sapiens')2.2 Test for gene set enrichment using the hypergeometric test
The hypergeometric test evaluates the over- or under-representation of a set of candidate genes in GO-categories, compared to a set of background genes (see Schematic 1 below).
The input for the hypergeometric test is a dataframe with two columns: (1) a column with gene-symbols and (2) a binary column with 1 for a candidate gene and 0 for a background gene.
2.2.1 Hypergeometric test using the default background gene set
The declaration of background genes is optional. If only candidate genes are defined, then all remaining genes from the annotation package are used as default background. In this example GO-enrichment of 13 human genes will be tested:
## load GOfuncR package
library(GOfuncR)
## create input dataframe with candidate genes
gene_ids = c('NCAPG', 'APOL4', 'NGFR', 'NXPH4', 'C21orf59', 'CACNG2', 'AGTR1',
'ANO1', 'BTBD3', 'MTUS1', 'CALB1', 'GYG1', 'PAX2')
input_hyper = data.frame(gene_ids, is_candidate=1)
input_hyper## gene_ids is_candidate
## 1 NCAPG 1
## 2 APOL4 1
## 3 NGFR 1
## 4 NXPH4 1
## 5 C21orf59 1
## 6 CACNG2 1
## 7 AGTR1 1
## 8 ANO1 1
## 9 BTBD3 1
## 10 MTUS1 1
## 11 CALB1 1
## 12 GYG1 1
## 13 PAX2 1This dataframe is the only mandatory input for go_enrich, however to lower computation time for the examples, we also lower the number of randomsets that are generated to compute the FWER:
## run enrichment analysis (n_randets=100 lowers compuation time
## compared to default 1000)
res_hyper = go_enrich(input_hyper, n_randset=100)The output of go_enrich is a list of 4 elements:
The most important is the first element which contains the results from the enrichment analysis (ordered by FWER for over-representation of candidate genes):
## first element of go_enrich result has the stats
stats = res_hyper[[1]]
## top-GO categories
head(stats)## ontology node_id node_name raw_p_underrep raw_p_overrep
## 1 biological_process GO:0072025 distal convoluted tubule development 1.000000 2.239363e-06
## 2 biological_process GO:0072221 metanephric distal convoluted tubule development 1.000000 2.239363e-06
## 3 biological_process GO:0072235 metanephric distal tubule development 1.000000 5.594437e-06
## 4 cellular_component GO:0098686 hippocampal mossy fiber to CA3 synapse 0.999999 1.908022e-04
## 5 biological_process GO:0072017 distal tubule development 1.000000 2.047659e-05
## 6 biological_process GO:0072205 metanephric collecting duct development 1.000000 2.047659e-05
## FWER_underrep FWER_overrep
## 1 1 0.00
## 2 1 0.00
## 3 1 0.01
## 4 1 0.01
## 5 1 0.03
## 6 1 0.03## top GO-categories per domain
by(stats, stats$ontology, head, n=3)## stats$ontology: biological_process
## ontology node_id node_name raw_p_underrep raw_p_overrep
## 1 biological_process GO:0072025 distal convoluted tubule development 1 2.239363e-06
## 2 biological_process GO:0072221 metanephric distal convoluted tubule development 1 2.239363e-06
## 3 biological_process GO:0072235 metanephric distal tubule development 1 5.594437e-06
## FWER_underrep FWER_overrep
## 1 1 0.00
## 2 1 0.00
## 3 1 0.01
## ----------------------------------------------------------------------------------
## stats$ontology: cellular_component
## ontology node_id node_name raw_p_underrep raw_p_overrep
## 4 cellular_component GO:0098686 hippocampal mossy fiber to CA3 synapse 0.9999990 0.0001908022
## 17 cellular_component GO:0098984 neuron to neuron synapse 0.9999572 0.0010757388
## 45 cellular_component GO:0044300 cerebellar mossy fiber 0.9999928 0.0042798122
## FWER_underrep FWER_overrep
## 4 1 0.01
## 17 1 0.22
## 45 1 0.80
## ----------------------------------------------------------------------------------
## stats$ontology: molecular_function
## ontology node_id
## 24 molecular_function GO:0001596
## 25 molecular_function GO:0099567
## 26 molecular_function GO:0008466
## node_name
## 24 angiotensin type I receptor activity
## 25 calcium ion binding involved in regulation of postsynaptic cytosolic calcium ion concentration
## 26 glycogenin glucosyltransferase activity
## raw_p_underrep raw_p_overrep FWER_underrep FWER_overrep
## 24 1.0000000 0.0006518551 1 0.53
## 25 1.0000000 0.0006518551 1 0.53
## 26 0.9999996 0.0013033206 1 0.69The second element is a dataframe with all valid input genes:
## all valid input genes
head(res_hyper[[2]])## gene_ids is_candidate
## 1 AGTR1 1
## 2 ANO1 1
## 3 APOL4 1
## 4 BTBD3 1
## 5 CACNG2 1
## 6 CALB1 1The third element states the reference genome for the annotations and the version of the GO-graph:
## annotation package used (default='Homo.sapiens') and GO-graph version
res_hyper[[3]]## type db version
## 1 go_annotations Homo.sapiens 1.3.1
## 2 go_graph integrated 23-Mar-2020The fourth element is a dataframe with the minimum p-values from the permutations, which are used to compute the FWER:
## minimum p-values from randomsets
head(res_hyper[[4]])## ontology lower_tail upper_tail
## 1 biological_process 0.0005317249 5.594437e-06
## 2 biological_process 0.0019038380 5.701863e-06
## 3 biological_process 0.0022349283 1.675952e-05
## 4 biological_process 0.0046521250 5.052538e-05
## 5 biological_process 0.0068259261 9.379225e-05
## 6 biological_process 0.0069524622 1.036803e-042.2.2 Hypergeometric test using a defined background gene set
Instead of using the default background gene set, it will often be more accurate to just include those genes in the background gene set, that were studied in the experiment that led to the discovery of the candidate genes.
For example, if the candidate genes are based on microarray expression data, than the background gene set should consist of all genes on the array.
To define a background gene set, just add lines to the input dataframe where the gene-associated variable in the second column is a 0.
Note that all candidate genes are implicitly part of the background gene set and do not need to be defined as background.
## create input dataframe with candidate and background genes
candi_gene_ids = c('NCAPG', 'APOL4', 'NGFR', 'NXPH4', 'C21orf59', 'CACNG2',
'AGTR1', 'ANO1', 'BTBD3', 'MTUS1', 'CALB1', 'GYG1', 'PAX2')
bg_gene_ids = c('FGR', 'NPHP1', 'DRD2', 'ABCC10', 'PTBP2', 'JPH4', 'SMARCC2',
'FN1', 'NODAL', 'CYP1A2', 'ACSS1', 'CDHR1', 'SLC25A36', 'LEPR', 'PRPS2',
'TNFAIP3', 'NKX3-1', 'LPAR2', 'PGAM2')
is_candidate = c(rep(1,length(candi_gene_ids)), rep(0,length(bg_gene_ids)))
input_hyper_bg = data.frame(gene_ids = c(candi_gene_ids, bg_gene_ids),
is_candidate)
head(input_hyper_bg)## gene_ids is_candidate
## 1 NCAPG 1
## 2 APOL4 1
## 3 NGFR 1
## 4 NXPH4 1
## 5 C21orf59 1
## 6 CACNG2 1tail(input_hyper_bg)## gene_ids is_candidate
## 27 LEPR 0
## 28 PRPS2 0
## 29 TNFAIP3 0
## 30 NKX3-1 0
## 31 LPAR2 0
## 32 PGAM2 0The enrichment analysis is performed like before, again with only 100 randomsets to lower computation time.
res_hyper_bg = go_enrich(input_hyper_bg, n_randsets=100)head(res_hyper_bg[[1]])## ontology node_id node_name raw_p_underrep raw_p_overrep FWER_underrep
## 1 cellular_component GO:0005634 nucleus 0.9848122 0.08044677 1
## 2 biological_process GO:0001678 cellular glucose homeostasis 1.0000000 0.04894327 1
## 3 biological_process GO:0006935 chemotaxis 1.0000000 0.04894327 1
## 4 biological_process GO:0042330 taxis 1.0000000 0.04894327 1
## 5 biological_process GO:1901653 cellular response to peptide 1.0000000 0.04894327 1
## 6 biological_process GO:0009605 response to external stimulus 0.9928047 0.05160760 1
## FWER_overrep
## 1 0.70
## 2 0.81
## 3 0.81
## 4 0.81
## 5 0.81
## 6 0.822.2.3 Hypergeometric test with correction for gene length
If the chance of a gene to be discovered as a candidate gene is higher for longer genes (e.g. the chance to have an amino-acid change compared to another species), it can be helpful to also correct for this length-bias in the calculation of the family-wise error rate (FWER).
go_enrich therefore offers the gene_len option: while with the default gene_len=FALSE candidate and background genes are permuted randomly in the randomsets (see Schematic 1), gene_len=TRUE makes the chance of a gene to be chosen as a candidate gene in a randomset dependent on its gene length.
## test input genes again with correction for gene length
res_hyper_len = go_enrich(input_hyper, gene_len=TRUE)Note that the default annotation package Homo.sapiens uses the hg19 gene-coordinates. See below for examples how to use other packages or custom gene-coordinates.
2.2.4 Hypergeometric test with genomic regions as input
Instead of defining candidate and background genes explicitly in the input dataframe, it is also possible to define entire chromosomal regions as candidate and background regions. The GO-enrichment is then tested for all genes located in, or overlapping the candidate regions on the plus or the minus strand.
In comparison to defining candidate and background genes explicitly, this option has the advantage that the FWER accounts for spatial clustering of genes.
For the random permutations used to compute the FWER, blocks as long as candidate regions are chosen from the merged candidate and background regions and genes contained in these blocks are considered candidate genes.
The option circ_chrom defines whether random candidate blocks are chosen from the same chromosome or not (Schematic 2).
To define chromosomal regions in the input dataframe, the entries in the first column have to be of the form chr:start-stop, where start always has to be smaller than stop.
Note that this option requires the input of background regions.
If multiple candidate regions are provided, in the randomsets they are placed randomly (but without overlap) into the merged candidate and background regions.
## create input vector with a candidate region on chromosome 8
## and background regions on chromosome 7, 8 and 9
regions = c('8:81000000-83000000', '7:1300000-56800000', '7:74900000-148700000',
'8:7400000-44300000', '8:47600000-146300000', '9:0-39200000',
'9:69700000-140200000')
is_candidate = c(1, rep(0,6))
input_regions = data.frame(regions, is_candidate)
input_regions## regions is_candidate
## 1 8:81000000-83000000 1
## 2 7:1300000-56800000 0
## 3 7:74900000-148700000 0
## 4 8:7400000-44300000 0
## 5 8:47600000-146300000 0
## 6 9:0-39200000 0
## 7 9:69700000-140200000 0## run GO-enrichment analysis for genes in the candidate region
res_region = go_enrich(input_regions, n_randsets=100, regions=TRUE)The output of go_enrich for genomic regions is identical to the one that is produced for single genes.
The first element of the output list contains the results of the enrichment analysis and the second element contains the candidate and background genes located in the user-defined regions:
stats_region = res_region[[1]]
head(stats_region)## ontology node_id node_name raw_p_underrep raw_p_overrep FWER_underrep
## 1 molecular_function GO:0005504 fatty acid binding 1 1.903569e-10 1.00
## 2 molecular_function GO:0033293 monocarboxylic acid binding 1 2.257592e-09 1.00
## 3 molecular_function GO:0031406 carboxylic acid binding 1 1.005555e-07 0.99
## 4 biological_process GO:0015908 fatty acid transport 1 2.538290e-08 0.99
## 5 biological_process GO:0015849 organic acid transport 1 4.974120e-08 0.99
## 6 biological_process GO:0006869 lipid transport 1 4.900076e-07 0.99
## FWER_overrep
## 1 0.06
## 2 0.06
## 3 0.07
## 4 0.09
## 5 0.09
## 6 0.09## see which genes are located in the candidate region
input_genes = res_region[[2]]
candidate_genes = input_genes[input_genes[,2]==1, 1]
candidate_genes## [1] "CHMP4C" "FABP4" "FABP5" "FABP9" "FABP12" "IMPA1" "PAG1" "PMP2" "SLC10A5" "SNX16"
## [11] "TPD52" "ZBTB10" "ZFAND1" "ZNF704"Note that the default annotation package Homo.sapiens uses the hg19 gene-coordinates. See below for examples how to use other packages or custom gene-coordinates.
2.3 Test for enrichment of high scored genes using the Wilcoxon rank-sum test
When genes are not divided into candidate and background genes, but are ranked by some kind of score, e.g. a p-value for differential expression, a Wilcoxon rank-sum test can be performed to find GO-categories where genes with high (or low) scores are over-represented. This example uses genes ranked by random scores:
## create input dataframe with scores in second column
high_score_genes = c('GCK', 'CALB1', 'PAX2', 'GYS1','SLC2A8', 'UGP2', 'BTBD3',
'MTUS1', 'SYP', 'PSEN1')
low_score_genes = c('CACNG2', 'ANO1', 'ZWINT', 'ENGASE', 'HK2', 'PYGL', 'GYG1')
gene_scores = c(runif(length(high_score_genes), 0.5, 1),
runif(length(low_score_genes), 0, 0.5))
input_willi = data.frame(gene_ids = c(high_score_genes, low_score_genes),
gene_scores)
head(input_willi)## gene_ids gene_scores
## 1 GCK 0.6284683
## 2 CALB1 0.7368831
## 3 PAX2 0.7648268
## 4 GYS1 0.7799225
## 5 SLC2A8 0.5804260
## 6 UGP2 0.5568016res_willi = go_enrich(input_willi, test='wilcoxon', n_randsets=100)The output is analogous to the one for the hypergeometric test:
head(res_willi[[1]])## ontology node_id node_name raw_p_low_rank raw_p_high_rank
## 1 biological_process GO:0032502 developmental process 0.9908011 0.01187070
## 2 biological_process GO:0000902 cell morphogenesis 0.9901175 0.01373432
## 3 biological_process GO:0000904 cell morphogenesis involved in differentiation 0.9901175 0.01373432
## 4 biological_process GO:0022008 neurogenesis 0.9901175 0.01373432
## 5 biological_process GO:0030030 cell projection organization 0.9901175 0.01373432
## 6 biological_process GO:0030154 cell differentiation 0.9901175 0.01373432
## FWER_low_rank FWER_high_rank
## 1 1 0.26
## 2 1 0.31
## 3 1 0.31
## 4 1 0.31
## 5 1 0.31
## 6 1 0.31Note that when p-values are used as scores, often one would want to look for enrichment of low ranks, i.e. low p-values (or alternatively use (1 - p-value) as score and check for enrichment of high ranks).
2.4 Test for enrichment using the binomial test
When genes are associated with two counts A and B, e.g. amino-acid changes since a common ancestor in two species, a binomial test can be used to identify GO-categories with an enrichment of genes with a high fraction of one of the counts compared to the fraction in the root node. To perform the binomial test the input dataframe needs a column with the gene symbols and two additional columns with the corresponding counts:
## create a toy example dataset with two counts per gene
high_A_genes = c('G6PD', 'GCK', 'GYS1', 'HK2', 'PYGL', 'SLC2A8', 'UGP2',
'ZWINT', 'ENGASE')
low_A_genes = c('CACNG2', 'AGTR1', 'ANO1', 'BTBD3', 'MTUS1', 'CALB1', 'GYG1',
'PAX2')
A_counts = c(sample(20:30, length(high_A_genes)),
sample(5:15, length(low_A_genes)))
B_counts = c(sample(5:15, length(high_A_genes)),
sample(20:30, length(low_A_genes)))
input_binom = data.frame(gene_ids=c(high_A_genes, low_A_genes), A_counts,
B_counts)
head(input_binom)## gene_ids A_counts B_counts
## 1 G6PD 29 9
## 2 GCK 20 8
## 3 GYS1 24 10
## 4 HK2 26 13
## 5 PYGL 23 5
## 6 SLC2A8 30 6In this example we also use the domains option to reduce the analysis to molecular_function and cellular_component.
Also the silent option is used, which suppresses all output that would be written to the screen (except for warnings and errors):
## run binomial test, excluding the 'biological_process' domain,
## suppress output to screen
res_binom = go_enrich(input_binom, test='binomial', n_randsets=100,
silent=TRUE, domains=c('molecular_function', 'cellular_component'))
head(res_binom[[1]])## ontology node_id node_name raw_p_high_B raw_p_high_A FWER_high_B
## 1 molecular_function GO:0005536 glucose binding 1 2.319179e-13 1
## 2 molecular_function GO:0030246 carbohydrate binding 1 2.319179e-13 1
## 3 molecular_function GO:0048029 monosaccharide binding 1 2.319179e-13 1
## 4 molecular_function GO:0036094 small molecule binding 1 3.533582e-09 1
## 5 molecular_function GO:0000166 nucleotide binding 1 4.926570e-09 1
## 6 molecular_function GO:1901265 nucleoside phosphate binding 1 4.926570e-09 1
## FWER_high_A
## 1 0.00
## 2 0.00
## 3 0.00
## 4 0.03
## 5 0.03
## 6 0.032.5 Test for enrichment using the 2x2 contingency table test
When genes are associated with four counts (A-D), e.g. non-synonymous or synonymous variants that are fixed between or variable within species, like for a McDonald-Kreitman test [3], the 2x2 contingency table test can be used. It can identify GO-categories which have a high ratio of A/B compared to C/D, which in this example would correspond to a high ratio of non-synonymous substitutions / synonymous substitutions compared to non-synonymous variable / synonymous variable:
## create a toy example with four counts per gene
high_substi_genes = c('G6PD', 'GCK', 'GYS1', 'HK2', 'PYGL', 'SLC2A8', 'UGP2',
'ZWINT', 'ENGASE')
low_substi_genes = c('CACNG2', 'AGTR1', 'ANO1', 'BTBD3', 'MTUS1', 'CALB1',
'GYG1', 'PAX2', 'C21orf59')
subs_non_syn = c(sample(5:15, length(high_substi_genes), replace=TRUE),
sample(0:5, length(low_substi_genes), replace=TRUE))
subs_syn = sample(5:15, length(c(high_substi_genes, low_substi_genes)),
replace=TRUE)
vari_non_syn = c(sample(0:5, length(high_substi_genes), replace=TRUE),
sample(0:10, length(low_substi_genes), replace=TRUE))
vari_syn = sample(5:15, length(c(high_substi_genes, low_substi_genes)),
replace=TRUE)
input_conti = data.frame(gene_ids=c(high_substi_genes, low_substi_genes),
subs_non_syn, subs_syn, vari_non_syn, vari_syn)
head(input_conti)## gene_ids subs_non_syn subs_syn vari_non_syn vari_syn
## 1 G6PD 9 6 4 8
## 2 GCK 10 12 0 11
## 3 GYS1 5 15 5 7
## 4 HK2 10 12 0 9
## 5 PYGL 5 13 3 12
## 6 SLC2A8 11 14 4 11## the corresponding contingency table for the first gene would be:
matrix(input_conti[1, 2:5], ncol=2,
dimnames=list(c('non-synonymous', 'synonymous'),
c('substitution','variable')))## substitution variable
## non-synonymous 9 4
## synonymous 6 8res_conti = go_enrich(input_conti, test='contingency', n_randset=100)The output is analogous to that of the other tests:
head(res_conti[[1]])## ontology node_id node_name raw_p_high_CD
## 1 molecular_function GO:0016772 transferase activity, transferring phosphorus-containing groups 1
## 2 molecular_function GO:0000166 nucleotide binding 1
## 3 molecular_function GO:1901265 nucleoside phosphate binding 1
## 4 molecular_function GO:0004340 glucokinase activity 1
## 5 molecular_function GO:0004396 hexokinase activity 1
## 6 molecular_function GO:0008865 fructokinase activity 1
## raw_p_high_AB FWER_high_CD FWER_high_AB
## 1 0.0001806997 1 0.08
## 2 0.0002031993 1 0.08
## 3 0.0002031993 1 0.08
## 4 0.0002560061 1 0.11
## 5 0.0002560061 1 0.11
## 6 0.0002560061 1 0.11Depending on the counts for each GO-category a Chi-square or Fisher’s exact test is performed. Note that this is the only test that is not dependent on the distribution of the gene-associated variables in the root nodes.
3 Enrichment analyses with different annotations or ontologies
3.1 Other annotation packages
Annotation package types suggested for GOfuncR:
| annotation package | information used in GOfuncR |
|---|---|
| OrganismDb | GO-annotations + gene-coordinates |
| OrgDb | GO-annotations |
| TxDb | gene-coordinates |
The default annotation package used by GOfuncR is bioconductor’s OrganismDb package Homo.sapiens, which contains GO-annotations as well as gene-coordinates.
There are currently also OrganismDb packages available for mouse (Mus.musculus) and rat (Rattus.norvegicus).
After installation those packages can be used in go_enrich:
## perform enrichment analysis for mouse genes
## ('Mus.musculus' has to be installed)
mouse_gene_ids = c('Gck', 'Gys1', 'Hk2', 'Pygl', 'Slc2a8', 'Ugp2', 'Zwint',
'Engase')
input_hyper_mouse = data.frame(mouse_gene_ids, is_candidate=1)
res_hyper_mouse = go_enrich(input_hyper_mouse, organismDb='Mus.musculus')Besides OrganismDb packages also OrgDb packages can be used to get GO-annotations.
These packages have the advantage that they are available for a broader range of species (e.g. org.Pt.eg.db for chimp or org.Gg.eg.db for chicken).
OrgDb packages are specified by the orgDb parameter of go_enrich:
## perform enrichment analysis for chimp genes
## ('org.Pt.eg.db' has to be installed)
chimp_gene_ids = c('SIAH1', 'MIIP', 'ELP3', 'CFB', 'ADARB1', 'TRNT1',
'DEFB124', 'OR1A1', 'TYR', 'HOXA7')
input_hyper_chimp = data.frame(chimp_gene_ids, is_candidate=1)
res_hyper_chimp = go_enrich(input_hyper_chimp, orgDb='org.Pt.eg.db')When an OrgDb package is used for annotations and the go_enrich analysis relies on gene-coordinates (i.e. gene_len=TRUE or regions=TRUE), then an additional TxDb package has to be provided for the gene-coordinates:
## perform enrichment analysis for chimp genes
## and account for gene-length in FWER
## (needs 'org.Pt.eg.db' and 'TxDb.Ptroglodytes.UCSC.panTro4.refGene' installed)
res_hyper_chimp_genelen = go_enrich(input_hyper_chimp, gene_len=TRUE,
orgDb='org.Pt.eg.db', txDb='TxDb.Ptroglodytes.UCSC.panTro4.refGene')OrgDb + TxDb packages can also be useful even if there is an OrganismDb package available, for example to use a different reference genome. Here we use the hg38 gene-coordinates from TxDb.Hsapiens.UCSC.hg38.knownGene instead of the default hg19 from the OrganismDb package Homo.sapiens.
## run GO-enrichment analysis for genes in the candidate region
## using hg38 gene-coordinates
## (needs 'org.Hs.eg.db' and 'TxDb.Hsapiens.UCSC.hg38.knownGene' installed)
res_region_hg38 = go_enrich(input_regions, regions=TRUE,
orgDb='org.Hs.eg.db', txDb='TxDb.Hsapiens.UCSC.hg38.knownGene')Note that using TxDb packages always requires defining an OrgDb package for the annotations.
3.2 Custom annotations
Besides using bioconductor’s annotation packages for the mapping of genes to GO-categories, it is also possible to provide the annotations directly as a dataframe with two columns: (1) genes and (2) GO-IDs (parameter annotations).
## example for a dataframe with custom annotations
head(custom_anno)## gene go_id
## 1 ABCC10 GO:0008559
## 2 ABCC10 GO:0015431
## 3 ABCC10 GO:0015698
## 4 ABCC10 GO:0016020
## 5 ABCC10 GO:0016021
## 6 ABCC10 GO:0016323## run enrichment analysis with custom annotations
res_hyper_anno = go_enrich(input_hyper, annotations=custom_anno)3.3 Custom gene-coordinates
Gene-coordinates are used when the FWER is corrected for gene length (gene_len=TRUE) or for spatial clustering of genes (regions=TRUE).
Instead of using gene-coordinates from bioconductor packages, one can also provide custom gene-coordinates directly as a dataframe with four columns: gene, chromosome, start, end (parameter gene_coords).
## example for a dataframe with custom gene-coordinates
head(custom_coords)## gene chr start end
## 1 NCAPG chr4 17812436 17846487
## 2 APOL4 chr22 36585176 36600879
## 3 NGFR chr17 47572655 47592382
## 4 NXPH4 chr12 57610578 57620232
## 5 C21orf59 chr21 33954510 33984918
## 6 CACNG2 chr22 36956916 37098690## use correction for gene-length based on custom gene-coordinates
res_hyper_cc = go_enrich(input_hyper, gene_len=TRUE, gene_coords=custom_coords)Note that this allows to use gene_len=TRUE to correct the FWER for any user-defined gene ‘weight’, since the correction for gene length just weights each gene with its length (end - start).
A gene with a higher weight has a bigger chance of becoming a candidate gene in the randomsets.
3.4 Custom GO-graph
A default GO-graph (obtained from geneontology, release date 01-May-2021), is integrated in the package.
However, also a custom GO-graph, e.g. a specific version or a different ontology can be provided. go_enrich needs a directory which contains three tab-separated files in the GO MySQL Database Schema:
term.txt, term2term.txt and graph_path.txt.
The full path to this directory needs to be defined in the parameter godir.
Specific versions of the GO-graph can be downloaded from http://archive.geneontology.org/lite/.
For example, to use the GO-graph from 2018-11-24, download and unpack the files from
http://archive.geneontology.org/lite/2018-11-24/go_weekly-termdb-tables.tar.gz.
Assume the files were saved in /home/user/go_graphs/2018-11-24/. This directory now contains the needed files term.txt, term2term.txt and graph_path.txt and can be used in go_enrich:
## run enrichment with custom GO-graph
go_path = '/home/user/go_graphs/2018-11-24/'
res_hyper = go_enrich(input_hyper, godir=go_path)3.4.1 Conversion from .obo format
At some point Gene Ontology may no longer provide the ontology in the GO MySQL Database Schema; and other ontologies may not be provided in that format at all.
Therefore, custom ontologies might need to be converted to the right format before using them in GOfuncR.
On https://github.com/sgrote/OboToTerm you can find a python script that converts the widely used .obo format to the tables needed (term.txt, term2term.txt and graph_path.txt).
4 Additional functionalities
4.1 Plot distribution of gene-associated variables from an enrichment analysis
The function plot_anno_scores can be used to get a quick visual overview of the gene-associated variables in GO-categories, that were used in an enrichment analysis.
plot_anno_scores takes a result from go_enrich as input together with a vector of GO-IDs.
It then plots the combined scores of all input genes for the go_enrich analysis in each of the defined GO-categories.
The type of the plot depends on the test that was used in go_enrich.
Note that if custom annotations were used in go_enrich, then they also have to be provided to plot_anno_scores (whereas ontology and annotation databases are inferred from the input and loaded in plot_anno_scores).
For the hypergeometric test pie charts show the amounts of candidate and background genes that are annotated to the GO-categories and the root nodes (candidate genes in the colour of the corresponding root node). The top panel shows the odds-ratio and 95%-CI from Fisher’s exact test (two-sided) comparing the GO-categories with their root nodes.
## hypergeometric test
top_gos_hyper = res_hyper[[1]][1:5, 'node_id']
# GO-categories with a high proportion of candidate genes
top_gos_hyper## [1] "GO:0072025" "GO:0072221" "GO:0072235" "GO:0098686" "GO:0072017"plot_anno_scores(res_hyper, top_gos_hyper)
plot_anno_scores returns an invisible dataframe that contains the stats from Fisher’s exact test shown in the plot:
## hypergeometric test with defined background
top_gos_hyper_bg = res_hyper_bg[[1]][1:5, 'node_id']
top_gos_hyper_bg## [1] "GO:0005634" "GO:0001678" "GO:0006935" "GO:0042330" "GO:1901653"plot_stats = plot_anno_scores(res_hyper_bg, top_gos_hyper_bg)
plot_stats## go_id candi_genes bg_genes root_id root_candi_genes root_bg_genes odds_ratio ci95_low ci95_high
## 1 GO:0005634 7 5 GO:0005575 12 19 3.734296 0.6672363 23.87889
## 2 GO:0001678 3 0 GO:0008150 12 19 Inf 0.7085715 Inf
## 3 GO:0006935 3 0 GO:0008150 12 19 Inf 0.7085715 Inf
## 4 GO:0042330 3 0 GO:0008150 12 19 Inf 0.7085715 Inf
## 5 GO:1901653 3 0 GO:0008150 12 19 Inf 0.7085715 Inf
## p
## 1 0.13043468
## 2 0.04894327
## 3 0.04894327
## 4 0.04894327
## 5 0.04894327Note that go_enrich reports the hypergeometric tests for over- and under-representation of candidate genes which correspond to the one-sided Fisher’s exact tests.
Also keep in mind that the p-values from this table are not corrected for multiple testing.
For the Wilcoxon rank-sum test violin plots show the distribution of the scores of genes that are annotated to each GO-category and the root nodes.
Horizontal lines in the left panel indicate the median of the scores that are annotated to the root nodes.
The Wilcoxon rank-sum test reported in the go_enrich result compares the scores annotated to a GO-category with the scores annotated to the corresponding root node.
## scores used for wilcoxon rank-sum test
top_gos_willi = res_willi[[1]][1:5, 'node_id']
# GO-categories with high scores
top_gos_willi## [1] "GO:0032502" "GO:0000902" "GO:0000904" "GO:0022008" "GO:0030030"plot_anno_scores(res_willi, top_gos_willi)
For the binomial test pie charts show the amounts of A and B counts associated with each GO-category and root node, (A in the colour of the corresponding root node).
The top-panel shows point estimates and the 95%-CI of p(A) in the nodes, as well as horizontal lines that correspond to p(A) in the root nodes.
The p-value in the returned object is based on the null hypothesis that p(A) in a node equals p(A) in the corresponding root node.
Note that go_enrich reports that value for one-sided binomial tests.
## counts used for the binomial test
top_gos_binom = res_binom[[1]][1:5, 'node_id']
# GO-categories with high proportion of A
top_gos_binom## [1] "GO:0005536" "GO:0030246" "GO:0048029" "GO:0036094" "GO:0000166"plot_anno_scores(res_binom, top_gos_binom)
Note that domain biological_process is missing in that plot because it was excluded from the GO-enrichment analysis in the first place (res_binom was created using the domains option of go_enrich).
For the 2x2 contingency table test pie charts show the proportions of A and B, as well as C and D counts associated with a GO-category.
Root nodes are not shown, because this test is independent of the root category.
The top panel shows the odds ratio and 95%-CI from Fisher’s exact test (two-sided) comparing A/B and C/D inside one node.
Note that in go_enrich, if all four values are >=10, a chi-square test is performed instead of Fisher’s exact test.
## counts used for the 2x2 contingency table test
top_gos_conti = res_conti[[1]][1:5, 'node_id']
# GO-categories with high A/B compared to C/D
top_gos_conti## [1] "GO:0016772" "GO:0000166" "GO:1901265" "GO:0004340" "GO:0004396"plot_anno_scores(res_conti, top_gos_conti)
4.2 Explore the GO-graph
The functions get_parent_nodes and get_child_nodes can be used to explore the ontology-graph.
They list all higher-level GO-categories and sub-GO-categories of input nodes, respectively, together with the distance between them:
## get the parent nodes (higher level GO-categories) of two GO-IDs
get_parent_nodes(c('GO:0051082', 'GO:0042254'))## child_go_id parent_go_id parent_name distance
## 1 GO:0042254 GO:0042254 ribosome biogenesis 0
## 2 GO:0042254 GO:0022613 ribonucleoprotein complex biogenesis 1
## 3 GO:0042254 GO:0044085 cellular component biogenesis 2
## 4 GO:0042254 GO:0071840 cellular component organization or biogenesis 3
## 5 GO:0042254 GO:0009987 cellular process 4
## 6 GO:0042254 GO:0008150 biological_process 5
## 7 GO:0051082 GO:0051082 unfolded protein binding 0
## 8 GO:0051082 GO:0005515 protein binding 1
## 9 GO:0051082 GO:0005488 binding 2
## 10 GO:0051082 GO:0003674 molecular_function 3## get the child nodes (sub-categories) of two GO-IDs
get_child_nodes(c('GO:0090070', 'GO:0000112'))## parent_go_id child_go_id child_name distance
## 1 GO:0000112 GO:0000112 nucleotide-excision repair factor 3 complex 0
## 2 GO:0000112 GO:0000440 core TFIIH complex portion of NEF3 complex 1
## 3 GO:0090070 GO:0090070 positive regulation of ribosome biogenesis 0Note that a GO-category per definition is also its own parent and child with distance 0.
The function get_names can be used to retrieve the names and root nodes of GO-IDs:
## get the full names and domains of two GO-IDs
get_names(c('GO:0090070', 'GO:0000112'))## go_id go_name root_node
## 1 GO:0090070 positive regulation of ribosome biogenesis biological_process
## 2 GO:0000112 nucleotide-excision repair factor 3 complex cellular_component It is also possible to go the other way round and search for GO-categories given part of their name using the function get_ids:
## get GO-IDs of categories that contain 'blood-brain barrier' in their names
bbb = get_ids('blood-brain barrier')
head(bbb)## node_name root_node go_id
## 1 establishment of endothelial blood-brain barrier biological_process GO:0014045
## 2 maintenance of blood-brain barrier biological_process GO:0035633
## 3 establishment of blood-brain barrier biological_process GO:0060856
## 4 establishment of glial blood-brain barrier biological_process GO:0060857
## 5 regulation of establishment of blood-brain barrier biological_process GO:0090210
## 6 positive regulation of establishment of blood-brain barrier biological_process GO:0090211Note that this is just a grep(..., ignore.case=TRUE) on the node names of the ontology.
More sophisticated searches, e.g. with regular expressions, could be performed on the table returned by get_ids('') which lists all non-obsolete GO-categories.
Like for go_enrich also custom ontologies can be used (see the help pages of the functions).
4.3 Retrieve associations between genes and GO-categories
GOfuncR also offers the functions get_anno_genes and get_anno_categories to get annotated genes given input GO-categories, and annotated GO-categories given input genes, respectively.
get_anno_genes takes a vector of GO-IDs as input, and returns all genes that are annotated to those categories (using the default annotation package Homo.sapiens).
The optional arguments database and genes can be used to define a different annotation package and the set of genes which is searched for annotations, respectively.
This function implicitly includes annotations to child nodes.
## find all genes that are annotated to GO:0000109
## using default database 'Homo.sapiens'
head(get_anno_genes(go_ids='GO:0000109'))## go_id gene
## 1 GO:0000109 CETN2
## 2 GO:0000109 ERCC1
## 3 GO:0000109 ERCC3
## 4 GO:0000109 ERCC4
## 5 GO:0000109 ERCC5
## 6 GO:0000109 ERCC8## find out wich genes from a set of genes
## are annotated to some GO-categories
genes = c('AGTR1', 'ANO1', 'CALB1', 'GYG1', 'PAX2')
gos = c('GO:0001558', 'GO:0005536', 'GO:0072205', 'GO:0006821')
anno_genes = get_anno_genes(go_ids=gos, genes=genes)
# add the names and domains of the GO-categories
cbind(anno_genes, get_names(anno_genes$go_id)[,2:3])## go_id gene go_name root_node
## 1 GO:0001558 AGTR1 regulation of cell growth biological_process
## 2 GO:0006821 ANO1 chloride transport biological_process
## 3 GO:0072205 CALB1 metanephric collecting duct development biological_process
## 4 GO:0072205 PAX2 metanephric collecting duct development biological_process## find all mouse-gene annotations to two GO-categories
## ('Mus.musculus' has to be installed)
gos = c('GO:0072205', 'GO:0000109')
get_anno_genes(go_ids=gos, database='Mus.musculus')get_anno_categories on the other hand uses gene-symbols as input and returns associated GO-categories:
## get the GO-annotations for two random genes
anno = get_anno_categories(c('BTC', 'SPAG5'))
head(anno)## gene go_id name domain
## 1 BTC GO:0005154 epidermal growth factor receptor binding molecular_function
## 2 BTC GO:0005515 protein binding molecular_function
## 3 BTC GO:0005576 extracellular region cellular_component
## 4 BTC GO:0005615 extracellular space cellular_component
## 5 BTC GO:0005886 plasma membrane cellular_component
## 6 BTC GO:0007173 epidermal growth factor receptor signaling pathway biological_process## get the GO-annotations for two mouse genes
## ('Mus.musculus' has to be installed)
anno_mus = get_anno_categories(c('Mus81', 'Papola'), database='Mus.musculus')This function only returns direct annotations.
To get also the parent nodes of the GO-categories a gene is annotated to, the function get_parent_nodes can be used:
# get all direct annotations of NXPH4
direct_anno = get_anno_categories('NXPH4')
direct_anno## gene go_id name domain
## 1 NXPH4 GO:0003674 molecular_function molecular_function
## 2 NXPH4 GO:0005102 signaling receptor binding molecular_function
## 3 NXPH4 GO:0005575 cellular_component cellular_component
## 4 NXPH4 GO:0005576 extracellular region cellular_component
## 5 NXPH4 GO:0007218 neuropeptide signaling pathway biological_process# get parent nodes of directly annotated GO-categories
parent_ids = unique(get_parent_nodes(direct_anno$go_id)[,2])
# add GO-domain
full_anno = get_names(parent_ids)
head(full_anno)## go_id go_name root_node
## 1 GO:0003674 molecular_function molecular_function
## 2 GO:0005102 signaling receptor binding molecular_function
## 3 GO:0005515 protein binding molecular_function
## 4 GO:0005488 binding molecular_function
## 5 GO:0005575 cellular_component cellular_component
## 6 GO:0005576 extracellular region cellular_componentLike for go_enrich also custom annotations and ontologies can be used (see the help pages of the functions).
4.4 Refine results from go_enrich
When there are many significant GO-categories given a FWER-threshold, it may be useful to restrict the results to the most specific categories.
The refine function implements the elim algorithm from [4], which removes genes from significant child categories and repeats the test to check whether a category would still be significant.
## perform enrichment analysis for some genes
gene_ids = c('NCAPG', 'APOL4', 'NGFR', 'NXPH4', 'C21orf59', 'CACNG2', 'AGTR1',
'ANO1', 'BTBD3', 'MTUS1', 'CALB1', 'GYG1', 'PAX2')
input_hyper = data.frame(gene_ids, is_candidate=1)
res_hyper = go_enrich(input_hyper, n_randset=100)
## perform refinement for categories with FWER < 0.1
refined = refine(res_hyper, fwer=0.1)## the result shows p-values and significance before and after refinement
refined## node_id ontology node_name raw_p_underrep raw_p_overrep
## 1 GO:0072025 biological_process distal convoluted tubule development 1.000000 2.239363e-06
## 2 GO:0072221 biological_process metanephric distal convoluted tubule development 1.000000 2.239363e-06
## 3 GO:0072235 biological_process metanephric distal tubule development 1.000000 5.594437e-06
## 4 GO:0098686 cellular_component hippocampal mossy fiber to CA3 synapse 0.999999 1.908022e-04
## 5 GO:0072017 biological_process distal tubule development 1.000000 2.047659e-05
## 6 GO:0072205 biological_process metanephric collecting duct development 1.000000 2.047659e-05
## FWER_underrep FWER_overrep refined_p_overrep signif
## 1 1 0.00 1.000000e+00 FALSE
## 2 1 0.00 2.239363e-06 TRUE
## 3 1 0.01 1.000000e+00 FALSE
## 4 1 0.03 1.908022e-04 TRUE
## 5 1 0.06 1.000000e+00 FALSE
## 6 1 0.06 2.047659e-05 TRUEBy default refine performs the test for over-representation of candidate genes, see ?refine for how to check for under-representation.
5 Schematics
5.1 Schematic 1: Hypergeometric test and FWER calculation
FWER calculation
The FWER for the other tests is computed in the same way: the gene-associated variables (scores or counts) are permuted while the annotations of genes to GO-categories stay fixed. Then the statistical tests are evaluated again for every GO-category.
5.2 Schematic 2: circ_chrom option for genomic regions input
options for genomic regions input
To use genomic regions as input, the first column of the genes input dataframe has to be of the form chr:start-stop and regions=TRUE has to be set.
The option circ_chrom defines how candidate regions are randomly moved inside the background regions for computing the FWER.
When circ_chrom=FALSE (default), candidate regions can be moved to any background region on any chromosome, but are not allowed to overlap multiple background regions.
When circ_chrom=TRUE, candidate regions are only moved on the same chromosome and are allowed to overlap multiple background regions.
The chromosome is ‘circularized’ which means that a randomly placed candidate region may start at the end of the chromosome and continue at the beginning.
6 Session Info
sessionInfo()## R version 4.2.0 RC (2022-04-19 r82224)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.4 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.15-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.15-bioc/R/lib/libRlapack.so
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C LC_TIME=en_GB
## [4] LC_COLLATE=C LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C LC_ADDRESS=C
## [10] LC_TELEPHONE=C LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] stats4 stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] Homo.sapiens_1.3.1 TxDb.Hsapiens.UCSC.hg19.knownGene_3.2.2
## [3] org.Hs.eg.db_3.15.0 GO.db_3.15.0
## [5] OrganismDbi_1.38.0 GenomicFeatures_1.48.0
## [7] GenomicRanges_1.48.0 GenomeInfoDb_1.32.0
## [9] AnnotationDbi_1.58.0 IRanges_2.30.0
## [11] S4Vectors_0.34.0 Biobase_2.56.0
## [13] BiocGenerics_0.42.0 GOfuncR_1.16.0
## [15] vioplot_0.3.7 zoo_1.8-10
## [17] sm_2.2-5.7 BiocStyle_2.24.0
##
## loaded via a namespace (and not attached):
## [1] bitops_1.0-7 matrixStats_0.62.0 bit64_4.0.5
## [4] filelock_1.0.2 progress_1.2.2 httr_1.4.2
## [7] tools_4.2.0 bslib_0.3.1 utf8_1.2.2
## [10] R6_2.5.1 DBI_1.1.2 tidyselect_1.1.2
## [13] prettyunits_1.1.1 bit_4.0.4 curl_4.3.2
## [16] compiler_4.2.0 graph_1.74.0 cli_3.3.0
## [19] xml2_1.3.3 DelayedArray_0.22.0 rtracklayer_1.56.0
## [22] bookdown_0.26 sass_0.4.1 RBGL_1.72.0
## [25] rappdirs_0.3.3 stringr_1.4.0 digest_0.6.29
## [28] Rsamtools_2.12.0 rmarkdown_2.14 XVector_0.36.0
## [31] pkgconfig_2.0.3 htmltools_0.5.2 MatrixGenerics_1.8.0
## [34] highr_0.9 dbplyr_2.1.1 fastmap_1.1.0
## [37] rlang_1.0.2 RSQLite_2.2.12 mapplots_1.5.1
## [40] jquerylib_0.1.4 BiocIO_1.6.0 generics_0.1.2
## [43] jsonlite_1.8.0 BiocParallel_1.30.0 gtools_3.9.2
## [46] dplyr_1.0.8 RCurl_1.98-1.6 magrittr_2.0.3
## [49] GenomeInfoDbData_1.2.8 Matrix_1.4-1 Rcpp_1.0.8.3
## [52] fansi_1.0.3 lifecycle_1.0.1 stringi_1.7.6
## [55] yaml_2.3.5 SummarizedExperiment_1.26.0 zlibbioc_1.42.0
## [58] BiocFileCache_2.4.0 grid_4.2.0 blob_1.2.3
## [61] parallel_4.2.0 crayon_1.5.1 lattice_0.20-45
## [64] Biostrings_2.64.0 hms_1.1.1 KEGGREST_1.36.0
## [67] magick_2.7.3 knitr_1.38 pillar_1.7.0
## [70] rjson_0.2.21 biomaRt_2.52.0 XML_3.99-0.9
## [73] glue_1.6.2 evaluate_0.15 BiocManager_1.30.17
## [76] png_0.1-7 vctrs_0.4.1 purrr_0.3.4
## [79] assertthat_0.2.1 cachem_1.0.6 xfun_0.30
## [82] restfulr_0.0.13 tibble_3.1.6 GenomicAlignments_1.32.0
## [85] memoise_2.0.1 ellipsis_0.3.27 References
[1] Ashburner, M. et al. (2000). Gene Ontology: tool for the unification of biology. Nature Genetics 25: 25-29. [https://doi.org/10.1038/75556]
[2] Pruefer, K. et al. (2007). FUNC: A package for detecting significant associations between gene sets and ontological annotations, BMC Bioinformatics 8: 41. [https://doi.org/10.1186/1471-2105-8-41]
[3] McDonald, J. H. Kreitman, M. (1991). Adaptive protein evolution at the Adh locus in Drosophila, Nature 351: 652-654. [https://doi.org/10.1038/351652a0]
[4] Alexa, A. et al. (2006). Improved scoring of functional groups from gene expression data by decorrelating GO graph structure. Bioinformatics 22, 1600–1607. [https://doi.org/10.1093/bioinformatics/btl140]