ABAEnrichment is designed to test user-defined genes for expression enrichment in different human brain regions.
The package integrates the expression of the input gene set and the structural information of the brain using an ontology, both provided by the Allen Brain Atlas project [1-4].
The statistical analysis is performed by the core function aba_enrich
which interfaces with the ontology enrichment software FUNC [5].
Additional functions provided in this package are get_expression
, plot_expression
, get_name
, get_id
, get_sampled_substructures
, get_superstructures
and get_annotated_genes
supporting the exploration and visualization of the expression data.
The package incorporates three different brain expression datasets:
All three datasets are filtered for protein-coding genes and gene expression is averaged across donors. Although the third dataset does not contain expression data, but a derived score, for simplicity we only refer to ‘expression’ in this documentation. For details on the datasets see the ABAData vignette.
Using the ontology that describes the hierarchical organization of the brain, brain regions get annotated all genes that are expressed in the brain region itself or in any of its substructures.
The boundary between ‘expressed’ and ‘not expressed’ is defined by different expression quantiles (e.g. using a quantile of 0.4, the lowest 40% of gene expression in the brain are considered ‘not expressed’ and the upper 60% are considered ‘expressed’).
These cutoffs are set with the parameter cutoff_quantiles
and an analysis is run for every cutoff separately.
The default cutoffs are 10% to 90% in steps of 10%.
The enrichment analysis is performed by using either the hypergeometric test, the Wilcoxon rank-sum test, the binomial test or the 2x2 contingency table test implemented in the ontology enrichment software FUNC [5]. The hypergeometric test evaluates the enrichment of annotated (expressed) candidate genes compared to annotated background genes for each brain region (see Schematic 1 below). The background genes can be defined explicitly like the candidate genes or, by default, consist of all protein-coding genes from the dataset that are not contained in the set of candidate genes. In contrast to this binary distinction between candidate and background genes, the Wilcoxon rank-sum test uses user-defined scores that are assigned to the input genes. It then tests every brain region for an enrichment of genes with high scores in the set of expressed input genes. 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 brain regions with an enrichment of expressed genes with a high fraction of A compared to the fraction of A in the brain in general. 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 [6], the 2x2 contingency table test can be used. It can identify brain regions which have a high ratio of A/B compared to C/D in their expressed genes.
To account for multiple testing, FUNC computes the family-wise error rate (FWER) using randomsets.
The randomsets are generated by permuting the gene-associated variables (e.g. candidate and background genes or the scores assigned to genes for the hypergeometric and Wilcoxon rank-sum test, respectively, see Schematic 1 below).
This is also the default behavior in ABAEnrichment.
For the hypergeometric test, ABAEnrichment additionally provides the option to correlate the chance of a background gene to be selected as a random candidate gene with the length of the background gene (option gene_len
).
Furthermore, instead of defining genes explicitly, whole genomic regions can be provided as input.
ABAEnrichment then tests brain regions for enrichment of expressed genes located in the candidate regions, compared to expressed genes located in the background regions.
The randomsets then also consist of randomly chosen candidate regions inside the background regions, either as a whole block in one background region (default), or on the same chromosome allowing to overlap multiple background regions on that chromosome (option circ_chrom
, see Schematic 2 below).
function | description |
---|---|
aba_enrich | core function for performing enrichment analyses given a candidate gene set |
get_expression | returns expression data for a given set of genes and brain regions |
plot_expression | plots a heatmap with expression data for a given set of genes and brain regions |
get_name | returns the full name of a brain region given a structure ID |
get_sampled_substructures | returns the substructures of a given brain region that have expression data available |
get_superstructures | returns the superstructures of a given brain region |
get_id | returns the structure ID given the name of a brain region |
get_annotated_genes | returns genes annotated to enriched or user-defined brain regions |
For a random set of 13 candidate genes, two analyses to identify human brain regions with enriched expression of the candidate genes are performed: one using data from adult donors (from Allen Human Brain Atlas [3]) and one using data from five developmental stages (from BrainSpan Atlas of the Developing Human Brain [4]).
The hypergeometric test evaluates the over-representation of a set of expressed candidate genes in brain regions, compared to a set of expressed background genes (see Schematic 1 below).
The input for the hypergeometric test is a dataframe with two columns: (1) a column with gene identifiers (Entrez-ID, Ensembl-ID or gene-symbol) and (2) a binary column with 1
for a candidate gene and 0
for a background gene.
In this example no background genes are defined, so all remaining protein-coding genes of the dataset are used as default background.
## load ABAEnrichment package
library(ABAEnrichment)
## create input vector with candidate genes
gene_ids = c('NCAPG', 'APOL4', 'NGFR', 'NXPH4', 'C21orf59', 'CACNG2', 'AGTR1',
'ANO1', 'BTBD3', 'MTUS1', 'CALB1', 'GYG1', 'PAX2')
is_candidate = rep(1, length(gene_ids))
input_hyper = data.frame(gene_ids, is_candidate)
head(input_hyper)
## gene_ids is_candidate
## 1 NCAPG 1
## 2 APOL4 1
## 3 NGFR 1
## 4 NXPH4 1
## 5 C21orf59 1
## 6 CACNG2 1
The core function aba_enrich
performs the enrichment analysis.
It takes the genes
vector as input, together with a dataset
argument which is set to adult
(default) or 5_stages
for the analyses of the adult and the developing human brain, respectively.
An example with the developmental effect score (dev_effect
) can be found below.
## run enrichment analyses with default parameters
## for the adult and developing human brain
res_adult = aba_enrich(input_hyper, dataset='adult')
res_devel = aba_enrich(input_hyper, dataset='5_stages')
In the following examples two additional parameters are set to lower computation time: cutoff_quantiles=c(0.5,0.7,0.9)
to use the 50%, 70% and 90% expression quantiles across all genes as the boundary between ‘expressed’ and ‘not expressed’ genes, and n_randsets=100
to use 100 random permutations to calculate the FWER.
cutoff_quantiles
and n_randsets
have default values seq(0.1,0.9,0.1)
and 1000
, respectively.
## run enrichment analysis with less cutoffs and randomsets
## to save computation time
res_devel = aba_enrich(input_hyper, dataset='5_stages',
cutoff_quantiles=c(0.5,0.7,0.9), n_randsets=100)
The function aba_enrich
returns a list, the first element of which contains the results of the statistical analysis for each brain region and age category (analyses are performed independently for each developmental stage):
## extract first element from the output list, which contains the statistics
fwers_devel = res_devel[[1]]
## see results for the brain regions with highest enrichment
## for children (3-11 yrs, age_category 3)
fwers_3 = fwers_devel[fwers_devel[,1]==3, ]
head(fwers_3)
## age_category structure_id structure n_significant mean_FWER min_FWER
## 55 3 Allen:10657 CBC_cerebellar cortex 0 0.5000000 0.36
## 56 3 Allen:10361 AMY_amygdaloid complex 0 0.9500000 0.85
## 57 3 Allen:10163 M1C_primary motor cortex (area M1, area 4) 0 0.9566667 0.87
## 58 3 Allen:10225 IPC_posteroventral (inferior) parietal cortex 0 0.9733333 0.92
## 59 3 Allen:10173 DFC_dorsolateral prefrontal cortex 0 0.9833333 0.96
## 60 3 Allen:10161 FCx_frontal neocortex 0 0.9866667 0.96
## equivalent_structures FWERs
## 55 Allen:10657;Allen:10656;Allen:10655;Allen:10654;Allen:10653 0.66;0.48;0.36
## 56 Allen:10361 0.85;1;1
## 57 Allen:10163;Allen:10162 0.87;1;1
## 58 Allen:10225;Allen:10214 0.92;1;1
## 59 Allen:10173 0.96;1;0.99
## 60 Allen:10161 0.96;1;1
The rows in the output data frame are ordered by age_category
, n_significant
, min_FWER
and mean_FWER
; with e.g. min_FWER
denoting the minimum FWER for enrichment of expressed candidate genes in that brain region across all expression cutoffs.
‘n_significant’ reports the number of cutoffs at which the FWER was below 0.05.
The column FWERs
lists the individual FWERs for each cutoff.
The column equivalent_structures
lists brain regions with identical expression data due to lack of independent expression measurements in all regions.
Nodes (brain regions) in the ontology inherit data from their children (substructures), and in the case of only one child node with expression data, the parent node inherits the child’s data leading to identical enrichment statistics.
In addition to the statistics, the list that is returned from aba_enrich
also contains the input genes for which expression data are available, and for each age category the gene expression values that correspond to the requested cutoff_quantiles
:
res_devel[2:3]
## $genes
## gene_ids is_candidate
## 1 AGTR1 1
## 2 ANO1 1
## 3 BTBD3 1
## 4 C21orf59 1
## 5 CACNG2 1
## 6 CALB1 1
## 7 GYG1 1
## 8 MTUS1 1
## 9 NCAPG 1
## 10 NGFR 1
## 11 NXPH4 1
## 12 PAX2 1
##
## $cutoffs
## age_category_1 age_category_2 age_category_3 age_category_4 age_category_5
## 50% 3.144481 2.854802 2.716617 2.776235 2.862117
## 70% 7.823920 7.017616 6.897414 6.842193 7.118609
## 90% 23.768641 22.478328 23.124388 21.625395 22.680811
For example, in the enrichment analysis of age category 2 (infant) with an expression cutoff of 0.7 (70%), genes are considered ‘expressed’ in a particular brain region when their expression value in that region is at least 7.017616.
The default behavior of aba_enrich
is to permute candidate and background genes randomly to compute the FWER.
With the option gene_len=TRUE
, random selection of background genes as candidate genes is dependent on the gene length, i.e. a gene twice as long as another gene also is twice as likely selected as a candidate gene in a randomset.
This is useful when the procedure that led to the identification of the candidate gene set is also more likely to discover longer genes.
Gene coordinates were obtained from http://grch37.ensembl.org/biomart/martview/ (GRCh37.p13).
The option ref_genome='grch38'
uses gene coordinates from the GRCh38 genome (GRCh38.p10) obtained from http://ensembl.org/biomart/martview/.
## run enrichment analysis, with randomsets dependent on gene length
res_len = aba_enrich(input_hyper, gene_len=TRUE)
## run the same analysis using gene coordinates
## from GRCh38 instead of the default GRCh37
res_len_grch38 = aba_enrich(input_hyper, gene_len=TRUE, ref_genome='grch38')
When the genes are not divided into candidate and background genes, but are ranked by scores, a Wilcoxon rank-sum test can be performed to find brain regions with a high proportion of genes with high scores in the set of expressed genes.
The second column of the genes
input dataframe then contains the scores assigned to the genes.
The output is identical to the one produced with the hypergeometric test.
## assign random scores to the genes used above
scores = sample(1:50, length(gene_ids))
input_wicox = data.frame(gene_ids, scores)
head(input_wicox)
## gene_ids scores
## 1 NCAPG 28
## 2 APOL4 21
## 3 NGFR 46
## 4 NXPH4 37
## 5 C21orf59 24
## 6 CACNG2 20
## test for enrichment of expressed genes with high scores in the adult brain
## using the Wilcoxon rank-sum test
res_wilcox = aba_enrich(input_wicox, test='wilcoxon',
cutoff_quantiles=c(0.5,0.7,0.9), n_randsets=100)
head(res_wilcox[[1]])
## age_category structure_id structure n_significant mean_FWER min_FWER
## 1 5 Allen:9017 ICjl_interstitial nucleus of Cajal, Right 0 0.8600000 0.59
## 2 5 Allen:4314 SI_Substantia Innominata, Right 0 0.9000000 0.74
## 3 5 Allen:12902 AHA_Anterior Hypothalamic Area 0 0.9033333 0.74
## 4 5 Allen:12919 VMH_Ventromedial Hypothalamic Nucleus 0 0.9033333 0.74
## 5 5 Allen:4598 AHA_Anterior Hypothalamic Area, Left 0 0.9033333 0.74
## 6 5 Allen:4627 LHA_Lateral hypothalamic area, Right 0 0.9033333 0.74
## equivalent_structures FWERs
## 1 Allen:9017 1;0.99;0.59
## 2 Allen:4314 0.97;0.99;0.74
## 3 Allen:12902 0.97;1;0.74
## 4 Allen:12919 0.97;1;0.74
## 5 Allen:4598 0.97;1;0.74
## 6 Allen:4627 0.97;1;0.74
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 brain regions with an enrichment of expressed genes with a high fraction A/(A+B) compared to the fraction of A in the brain in general (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('RFFL', 'NTS', 'LIPE', 'GALNT6', 'GSN', 'BTBD16', 'CERS2')
low_A_genes = c('GDA', 'ENC1', 'EGR4', 'VIPR1', 'DOC2A', 'OASL', 'FRY', 'NAV3')
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 RFFL 27 13
## 2 NTS 21 8
## 3 LIPE 30 7
## 4 GALNT6 29 11
## 5 GSN 28 6
## 6 BTBD16 25 9
In this example 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
res_binom = aba_enrich(input_binom, cutoff_quantiles=c(0.2,0.9),
test='binomial', n_randsets=100, silent=TRUE)
head(res_binom[[1]])
## age_category structure_id structure n_significant mean_FWER
## 1 5 Allen:4518 Sb_Subthalamic Nucleus, Left 0 0.125
## 2 5 Allen:12923 DTL_Lateral Group of Nuclei 0 0.185
## 3 5 Allen:12925 DTLv_Lateral Group of Nuclei, Ventral Division 0 0.185
## 4 5 Allen:4417 DTLv_Lateral group of Nuclei, Left, Ventral division 0 0.185
## 5 5 Allen:9054 RN_Red Nucleus, Left 0 0.185
## 6 5 Allen:9598 GiRt_gigantocellular group, Left 0 0.185
## min_FWER equivalent_structures FWERs
## 1 0.12 Allen:4518 0.12;0.13
## 2 0.12 Allen:12923 0.12;0.25
## 3 0.12 Allen:12925 0.12;0.25
## 4 0.12 Allen:4417 0.12;0.25
## 5 0.12 Allen:9054 0.12;0.25
## 6 0.12 Allen:9598 0.12;0.25
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 [6], the 2x2 contingency table test can be used. It can identify brain regions 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('RFFL', 'NTS', 'LIPE', 'GALNT6', 'GSN', 'BTBD16', 'CERS2')
low_substi_genes = c('ENC1', 'EGR4', 'NPTX1', 'DOC2A', 'OASL', 'FRY', 'NAV3')
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 RFFL 9 5 2 15
## 2 NTS 15 6 2 5
## 3 LIPE 13 14 4 6
## 4 GALNT6 11 10 5 8
## 5 GSN 12 13 4 15
## 6 BTBD16 11 10 1 10
## 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 2
## synonymous 5 15
res_conti = aba_enrich(input_conti, test='contingency',
cutoff_quantiles=c(0.7,0.8,0.9), n_randset=100)
The output is analogous to that of the other tests:
head(res_conti[[1]])
## age_category structure_id structure n_significant mean_FWER min_FWER equivalent_structures
## 1 5 Allen:4671 MB_Mammillary Body, Left 1 0.6700000 0.01 Allen:4671
## 2 5 Allen:9512 MY_Myelencephalon 1 0.3833333 0.04 Allen:9512
## 3 5 Allen:4391 DiE_Diencephalon 1 0.4900000 0.04 Allen:4391
## 4 5 Allen:4392 TH_Thalamus 1 0.4900000 0.04 Allen:4392
## 5 5 Allen:4665 MamR_Mammillary Region 1 0.5000000 0.04 Allen:4665
## 6 5 Allen:12909 MB_Mammillary Body 1 0.6800000 0.04 Allen:12909
## FWERs
## 1 0.01;1;1
## 2 0.04;0.11;1
## 3 0.43;0.04;1
## 4 0.43;0.04;1
## 5 0.04;0.46;1
## 6 0.04;1;1
Depending 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.
Instead of defining candidate and background genes explicitly in the genes
input dataframe, it is also possible to define entire chromosomal regions as candidate and background regions.
The expression enrichment is then tested for all protein-coding genes located in, or overlapping the candidate regions on the plus or the minus strand.
The gene coordinates used to identify those genes were obtained from http://grch37.ensembl.org/biomart/martview/ (GRCh37.p13).
The option ref_genome='grch38'
uses gene coordinates from the GRCh38.p10 genome version obtained from http://ensembl.org/biomart/martview/.
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 (Schematic 2).
To define chromosomal regions in the input dataframe, the first column has 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.
The output of aba_enrich
is identical to the one that is produced for single genes.
The second element of the output list contains the candidate and background genes located in the user-defined regions:
## create input vector with a candidate region on chromosome 8
## and background regions on chromosome 7, 8 and 9
regions = c('8:82000000-83000000', '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_region = data.frame(regions, is_candidate)
## run enrichment analysis for the adult human brain
res_region = aba_enrich(input_region, dataset='adult',
cutoff_quantiles=c(0.5,0.7,0.9), n_randsets=100)
## look at the results from the enrichment analysis
fwers_region = res_region[[1]]
head(fwers_region)
## age_category structure_id structure n_significant mean_FWER min_FWER
## 1 5 Allen:12926 MG_Medial Geniculate Complex 1 0.5166667 0.01
## 2 5 Allen:9150 LC_locus ceruleus, Right 1 0.1566667 0.02
## 3 5 Allen:4671 MB_Mammillary Body, Left 1 0.3833333 0.02
## 4 5 Allen:4734 He-III_III, Left Lateral Hemisphere 1 0.3366667 0.03
## 5 5 Allen:4738 He-VI_VI, Left Lateral Hemisphere 1 0.3433333 0.03
## 6 5 Allen:12909 MB_Mammillary Body 1 0.4100000 0.03
## equivalent_structures FWERs
## 1 Allen:12926 0.78;0.76;0.01
## 2 Allen:9150 0.3;0.02;0.15
## 3 Allen:4671 0.78;0.35;0.02
## 4 Allen:4734 0.53;0.45;0.03
## 5 Allen:4738 0.54;0.46;0.03
## 6 Allen:12909 0.78;0.42;0.03
## see which genes are located in the candidate region
input_genes = res_region[[2]]
candidate_genes = input_genes[input_genes[,2]==1,]
candidate_genes
## gene score
## 278 CHMP4C 1
## 484 FABP4 1
## 485 FABP5 1
## 486 FABP9 1
## 487 FABP12 1
## 727 IMPA1 1
## 1053 PAG1 1
## 1117 PMP2 1
## 1347 SLC10A5 1
## 1393 SNX16 1
## 1691 ZFAND1 1
An alternative method to choose random blocks from the background regions can be used with the option circ_chrom=TRUE
.
Every candidate region is then compared to background regions on the same chromosome (Schematic 2).
And in contrast to the default circ_chrom=FALSE
, randomly chosen blocks do not have to be located inside a single background region, but are allowed to overlap multiple background regions.
This means that a randomly chosen block can start at the end of the last background region and continue at the beginning of the first background region on a given chromosome.
The function get_expression
enables the output of gene and brain region-specific expression data averaged across donors.
By only setting the parameter structure_ids
that defines the brain regions, the gene_ids
and dataset
are automatically set to the genes and dataset used in the last enrichment analysis.
In comparison to defining genes and brain regions explicitly this saves some time since some pre-computations on the original dataset, e.g. aggregation of expression per gene, do not have to be redone.
Using the default options (background=FALSE
), get_expression
returns expression data for the candidate genes.
If background=TRUE
, the gene expression data for both, candidate genes and background genes, are returned.
## get expression data for the first 5 brain regions
## from the last aba_enrich-analysis
top_regions = fwers_region[1:5, 'structure_id']
top_regions
## [1] "Allen:12926" "Allen:9150" "Allen:4671" "Allen:4734" "Allen:4738"
expr = get_expression(top_regions, background=FALSE)
head(expr)
## CHMP4C FABP12 FABP4 FABP5 FABP9 IMPA1 PAG1 PMP2 SLC10A5 SNX16 ZFAND1
## Allen:4444 2.586348 1.648789 2.129794 7.979775 1.358975 9.586679 8.861224 11.05810 1.990756 6.375037 8.401070
## Allen:4499 3.086266 1.302643 3.250694 9.242289 1.293894 8.830867 8.682235 11.50669 1.704897 6.400059 8.977492
## Allen:9150 2.801755 1.737777 2.188948 9.165938 1.568329 8.782606 7.299574 10.75592 1.969351 7.270022 8.160494
## Allen:4671 3.315379 1.322007 2.948199 8.331502 1.289105 9.367726 8.846593 10.50742 2.517208 6.726171 8.877031
## Allen:4675 2.784213 1.969504 3.043131 9.223920 1.588357 8.954029 7.125901 10.47701 2.137033 7.080715 8.722243
## Allen:4672 2.645451 1.744787 2.177720 7.968980 1.448837 9.034155 8.390096 10.48329 2.437719 6.958238 8.863811
The same output would be created independently of an aba_enrich
analysis by, in addition to structure_ids
, setting gene_ids
and dataset
manually.
Like in all functions of the ABAEnrichment package gene_ids
can be Entrez-ID, Ensembl-ID or gene-symbol.
## get expression data independent from previous aba_enrich analysis
regions = c('Allen:12926', 'Allen:4738', 'Allen:4671', 'Allen:12909')
gene_ids = c('CHMP4C', 'FABP12', 'FABP4', 'FABP5', 'FABP9', 'IMPA1',
'PAG1', 'PMP2', 'SLC10A5', 'SNX16', 'ZFAND1')
expr2 = get_expression(regions, gene_ids=gene_ids, dataset='adult',
background=FALSE)
For the 5_stages
dataset the output of get_expression
is a list with a data frame for each developmental stage, where the first element corresponds to the first developmental stage, the second element to the second developmental stage, and so on.
Note that the brain regions passed to get_expression
do not have to match the brain regions returned in the output.
This is due to the fact that not all brain regions were measured independently.
In case a brain region was not measured directly, all available expression data from its substructures are returned.
The function get_sampled_substructures can be used to identify substructures with expression data.
The function plot_expression
enables the visualization of expression data.
The usage of plot_expression
is similar to that of get_expression
.
Providing only brain regions as input, it plots the expression data for the genes and dataset used in the last aba_enrich
call.
## get expression data for the first 5 brain regions
## from the last aba_enrich-analysis
top_regions = fwers_region[1:5, 'structure_id']
plot_expression(top_regions, background=FALSE)
The optional argument dendro
determines whether or not a dendrogram should be added to the heatplot.
The colored side bar in the plot without dendrogram indicates candidate genes (red) and background genes (black).
In this case only candidate gene expression was plotted (with the default option background=FALSE
):
## plot the same expression data without dendrogram
plot_expression(top_regions, dendro=FALSE, background=FALSE)
When plotting expression data following an enrichment analysis with the Wilcoxon rank-sum test, the option dendro=FALSE
results in a side bar that indicates the scores that were used for the enrichment analysis. For the binomial test the side bar shows A/(A+B) and for the 2x2 contingency table test ((A+1)/(B+1)) / ((C+1)/(D+1)) (+1 added to prevent division by 0, this is just a visual indication of the proportion of the ratios and not the real odds ratio from the 2x2 contingency table test).
Like get_expression
, plot_expression
can also be used independently of an enrichment analysis.
In that case the arguments gene_ids
and dataset
have to be defined.
If the 5_stages
dataset is used, the additional argument age_category
selects the developmental stage for which the expression data should be plotted:
## plot expression of some genes for the frontal neocortex (Allen:10161)
## in age category 3 (children, 3-11 yrs)
gene_ids = c('ENSG00000157764', 'ENSG00000163041', 'ENSG00000182158',
'ENSG00000147889', 'ENSG00000103126', 'ENSG00000184634')
plot_expression('Allen:10161', gene_ids=gene_ids, dataset='5_stages',
age_category=3)
In this example the frontal neocortex (Allen:10161) was not sampled directly for expression measurements, but some of its substructures have expression data annotated. Instead of pooling the data of the substructures, they are plotted separately.
As illustrated in the previous example, some brain regions like frontal neocortex (Allen:10161) do not have gene expression data available in the data set, but some of their substructures do have. Plotting or requesting expression data for such brain regions automatically obtains the data for all its sampled substructures.
ABAEnrichment offers some functions to explore the ontology graph which describes the hierarchical organization of the brain regions used in the enrichment analyses.
The function get_sampled_substructures
returns the IDs of all the substructures for which expression data are available, and get_superstructures
returns all superstructures in the order ‘general to special’.
The function get_name
is useful to see the name of a brain region given a structure ID:
## get IDs of the substructures of the frontal neocortex (Allen:10161)
## for which expression data are available
subs = get_sampled_substructures('Allen:10161')
subs
## [1] "Allen:10173" "Allen:10185" "Allen:10194" "Allen:10163"
## get the full name of those substructures
get_name(subs)
## Allen:10173 Allen:10185
## "DFC_dorsolateral prefrontal cortex" "VFC_ventrolateral prefrontal cortex"
## Allen:10194 Allen:10163
## "OFC_orbital frontal cortex" "M1C_primary motor cortex (area M1, area 4)"
## get the superstructures of the frontal neocortex (from general to special)
supers = get_superstructures('Allen:10161')
supers
## [1] "Allen:10153" "Allen:10154" "Allen:10155" "Allen:10156" "Allen:10157" "Allen:10158" "Allen:10159"
## [8] "Allen:10160" "Allen:10161"
## get the full names of those superstructures
get_name(supers)
## Allen:10153 Allen:10154 Allen:10155
## "NP_neural plate" "NT_neural tube" "Br_brain"
## Allen:10156 Allen:10157 Allen:10158
## "F_forebrain (prosencephalon)" "FGM_gray matter of forebrain" "Tel_telencephalon"
## Allen:10159 Allen:10160 Allen:10161
## "Cx_cerebral cortex" "NCx_neocortex (isocortex)" "FCx_frontal neocortex"
Note that the ontologies and the IDs for brain regions differ between the adult and the developing brain.
However, the ontology functions get_name
, get_sampled_substructures
and get_superstructures
work with valid brain regions IDs from both ontologies.
The function get_id
searches the ontologies of the adult and developing brain for the structure ID that belongs to a given brain region name:
## get structure IDs of brain regions that contain 'accumbens' in their name
get_id('accumbens')
## structure ontology structure_id
## 1 Acb_Nucleus Accumbens adult Allen:4290
## 2 Acb_Nucleus Accumbens, Left adult Allen:4291
## 3 Acb_Nucleus Accumbens, Right adult Allen:4292
## get structure IDs of brain regions that contain 'telencephalon' in their name
get_id('telencephalon')
## structure ontology structure_id
## 1 Tel_telencephalon developmental Allen:10158
## 2 Tel_Telencephalon adult Allen:4007
Note that the output of get_id
is restricted to brain regions that are used in ABAEnrichment.
The function can also be used to get a full list of covered brain regions together with their IDs:
## get all brain regions included in ABAEnrichment together with thier IDs
all_regions = get_id('')
head(all_regions)
## structure ontology structure_id
## 1 NP_neural plate developmental Allen:10153
## 2 NT_neural tube developmental Allen:10154
## 3 Br_brain developmental Allen:10155
## 4 F_forebrain (prosencephalon) developmental Allen:10156
## 5 FGM_gray matter of forebrain developmental Allen:10157
## 6 Tel_telencephalon developmental Allen:10158
Often it is useful to see which genes are annotated to a brain region, i.e. count as ‘expressed’, given an expression cutoff.
While this can be accomplished using the expression cutoffs from aba_enrich(...)[[3]]
and the expression values from get_expression
, ABAEnrichment now also offers the convenient function get_annotated_genes
.
This function takes the output from aba_enrich
and a FWER-threshold (fwer_threshold
, default=0.05) as input and returns all brain-region/expression-cutoff combinations with a FWER below the FWER-threshold together with the corresponding annotated genes, the FWER and the score (1 for candidate and 0 for background genes or the scores from the Wilcoxon rank-sum test).
Note that a brain region gets all genes annotated that are expressed in the brain region itself or in any of the sampled substructures (see Schematic 1 below).
## run an enrichment analysis with 7 candidate and 7 background genes
## for the developing brain
gene_ids = c('FOXJ1', 'NTS', 'LTBP1', 'STON2', 'KCNJ6', 'AGT',
'ANO3', 'TTR', 'ELAVL4', 'BEAN1', 'TOX', 'EPN3', 'PAX2', 'KLHL1')
is_candidate = rep(c(1,0), each=7)
input_hyper = data.frame(gene_ids, is_candidate)
res = aba_enrich(input_hyper, dataset='5_stages',
cutoff_quantiles=c(0.5,0.7,0.9), n_randsets=100)
head(res[[1]])
## age_category structure_id structure n_significant mean_FWER min_FWER
## 1 1 Allen:10294 HIP_hippocampus (hippocampal formation) 0 0.4666667 0.25
## 2 1 Allen:10398 MD_mediodorsal nucleus of thalamus 0 0.7800000 0.47
## 3 1 Allen:10331 CN_cerebral nuclei 0 0.9066667 0.72
## 4 1 Allen:10361 AMY_amygdaloid complex 0 0.9066667 0.72
## 5 1 Allen:10157 FGM_gray matter of forebrain 0 0.9433333 0.83
## 6 1 Allen:10158 Tel_telencephalon 0 0.9433333 0.83
## equivalent_structures FWERs
## 1 Allen:10294;Allen:10293;Allen:10292 0.72;0.43;0.25
## 2 Allen:10398;Allen:10397;Allen:10391;Allen:10390;Allen:10389 0.87;1;0.47
## 3 Allen:10331 0.72;1;1
## 4 Allen:10361 0.72;1;1
## 5 Allen:10157;Allen:10156 0.83;1;1
## 6 Allen:10158 0.83;1;1
## see which candidate genes are annotated to brain regions with a FWER < 0.01
anno = get_annotated_genes(res, fwer_threshold=0.1)
anno
## age_category structure_id cutoff FWER anno_gene score
## 1 5 Allen:10333 0.5 0.08 AGT 1
## 2 5 Allen:10333 0.5 0.08 ANO3 1
## 3 5 Allen:10333 0.5 0.08 LTBP1 1
## 4 5 Allen:10333 0.5 0.08 STON2 1
In addition to passing the result of an enrichment analysis together with a FWER-threshold to get_annotated_genes
, it is also possible to define brain regions, expression cutoffs and (optionally) genes directly.
The function then returns all genes that have expression above the cutoffs in the respective brain regions.
If genes
are defined, the output is reduced to this set of genes; if not, all protein-coding genes with available expression data are analyzed.
## find out which of the above genes have expression
## above the 70% and 90% expression-cutoff
## in the basal ganglia of the adult human brain (Allen:4276)
anno2 = get_annotated_genes(structure_ids='Allen:4276', dataset='adult',
cutoff_quantiles=c(0.7,0.9), genes=gene_ids)
anno2
## age_category structure_id cutoff anno_gene
## 1 5 Allen:4276 0.7 AGT
## 2 5 Allen:4276 0.7 ANO3
## 3 5 Allen:4276 0.7 TTR
## 4 5 Allen:4276 0.9 AGT
## 5 5 Allen:4276 0.9 ANO3
## 6 5 Allen:4276 0.9 TTR
In the previous examples genes got annotated to brain regions based on their expression.
Besides the two gene expression datasets adult
and 5_stages
, the dataset dev_effect
can be used, which provides scores for an age effect for genes based on their expression change during development.
Using this dataset, the same analyses as above are performed, except that a gene is annotated to a brain region when its developmental effect score in that region is a above the cutoff_quantiles
.
To test brain regions for enrichment of candidate genes in the set of all genes with a developmental effect score above cutoff, the dataset
parameter has to be set to dev_effect
.
The output of the developmental effect enrichment analysis is equal to that of the expression enrichment analysis:
## use previously defined input genes dataframe
head(input_hyper)
## gene_ids is_candidate
## 1 FOXJ1 1
## 2 NTS 1
## 3 LTBP1 1
## 4 STON2 1
## 5 KCNJ6 1
## 6 AGT 1
## run enrichment analysis with developmental effect score
res_dev_effect = aba_enrich(input_hyper, dataset='dev_effect',
cutoff_quantiles=c(0.5,0.7,0.9), n_randsets=100)
## see the 5 brain regions with the lowest FWERs
top_regions = res_dev_effect[[1]][1:5, ]
top_regions
## age_category structure_id structure n_significant
## 1 0 Allen:10294 HIP_hippocampus (hippocampal formation) 0
## 2 0 Allen:10269 V1C_primary visual cortex (striate cortex, area V1/17) 0
## 3 0 Allen:10235 TCx_temporal neocortex 0
## 4 0 Allen:10243 STC_posterior (caudal) superior temporal cortex (area 22c) 0
## 5 0 Allen:13322 DLTC_dorsolateral temporal neocortex 0
## mean_FWER min_FWER equivalent_structures FWERs
## 1 0.7200000 0.67 Allen:10294;Allen:10293;Allen:10292 0.67;0.72;0.77
## 2 0.7933333 0.67 Allen:10269;Allen:10268 0.67;0.72;0.99
## 3 0.8333333 0.73 Allen:10235 1;0.73;0.77
## 4 0.8333333 0.73 Allen:10243;Allen:10240 1;0.73;0.77
## 5 0.8333333 0.73 Allen:13322 1;0.73;0.77
As for the expression datasets, the developmental effect scores can be retrieved with the functions get_expression
and plotted with plot_expression
:
## plot developmental effect score of the 5 brain regions with the lowest FWERs
plot_expression(top_regions[ ,'structure_id'])
The FWERs for the other three tests are computed analogously: first, for every brain region a statistical test is performed to get an enrichment p-value, then the scores or counts that are assigned to the genes in the input data are permuted and p-values are recomputed, and finally the original p-values are compared to the minimum p-values from the randomsets.
circ_chrom
option for genomic regions inputTo use genomic regions as input, the first column of the genes
input dataframe has to be of the form chr:start-stop
.
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.
sessionInfo()
## R version 3.5.0 (2018-04-23)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.4 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.7-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.7-bioc/R/lib/libRlapack.so
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C LC_TIME=en_US.UTF-8
## [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] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] ABAEnrichment_1.10.0 BiocStyle_2.8.0
##
## loaded via a namespace (and not attached):
## [1] Rcpp_0.12.16 bookdown_0.7 gtools_3.5.0 digest_0.6.15 rprojroot_1.3-2
## [6] bitops_1.0-6 backports_1.1.2 magrittr_1.5 evaluate_0.10.1 KernSmooth_2.23-15
## [11] gplots_3.0.1 stringi_1.1.7 data.table_1.10.4-3 gdata_2.18.0 rmarkdown_1.9
## [16] tools_3.5.0 stringr_1.3.0 ABAData_1.9.0 xfun_0.1 yaml_2.1.18
## [21] compiler_3.5.0 caTools_1.17.1 htmltools_0.3.6 knitr_1.20
[1] Hawrylycz, M.J. et al. (2012) An anatomically comprehensive atlas of the adult human brain transcriptome, Nature 489: 391-399. [doi:10.1038/nature11405]
[2] Miller, J.A. et al. (2014) Transcriptional landscape of the prenatal human brain, Nature 508: 199-206. [doi:10.1038/nature13185]
[3] Allen Institute for Brain Science. Allen Human Brain Atlas (Internet). Available from: [http://human.brain-map.org/]
[4] Allen Institute for Brain Science. BrainSpan Atlas of the Developing Human Brain (Internet). Available from: [http://brainspan.org/]
[5] Pruefer, K. et al. (2007) FUNC: A package for detecting significant associations between gene sets and ontological annotations, BMC Bioinformatics 8: 41. [doi:10.1186/1471-2105-8-41]
[6] McDonald, J. H. Kreitman, M. (1991). Adaptive protein evolution at the Adh locus in Drosophila, Nature 351: 652-654. [doi:10.1038/351652a0]