Contents

1 Overview

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.

1.1 Expression data

The package incorporates three different brain expression datasets:

  1. microarray data from six adult individuals
  2. RNA-seq data from 42 individuals of five different developmental stages (prenatal, infant, child, adolescent, adult)
  3. developmental effect scores measuring the age effect on expression per gene

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.

1.2 Annotation of genes to brain regions

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%.

1.3 Enrichment analysis

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.

input data and test selection

input data and test selection

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).

1.4 Functions included in ABAEnrichment

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 given a matrix of expression data
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

2 Examples

2.1 Test for gene expression enrichment using the hypergeometric test

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 data.frame 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)
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
## 2             3  Allen:10657                         CBC_cerebellar cortex             0 0.5000000     0.36
## 8             3  Allen:10361                        AMY_amygdaloid complex             0 0.9500000     0.85
## 13            3  Allen:10163    M1C_primary motor cortex (area M1, area 4)             0 0.9566667     0.87
## 20            3  Allen:10225 IPC_posteroventral (inferior) parietal cortex             0 0.9733333     0.92
## 34            3  Allen:10173            DFC_dorsolateral prefrontal cortex             0 0.9833333     0.96
## 36            3  Allen:10161                         FCx_frontal neocortex             0 0.9866667     0.96
##                                          equivalent_structures          FWERs
## 2  Allen:10657;Allen:10656;Allen:10655;Allen:10654;Allen:10653 0.66;0.48;0.36
## 8                                                  Allen:10361       0.85;1;1
## 13                                     Allen:10163;Allen:10162       0.87;1;1
## 20                                     Allen:10225;Allen:10214       0.92;1;1
## 34                                                 Allen:10173    0.96;1;0.99
## 36                                                 Allen:10161       0.96;1;1

The rows in the output data frame are ordered by n_significant, min_FWER and mean_FWER, age_category and structure_id; 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.

2.1.1 Correct for gene length

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/. Alternatively also custom gene-coordinates can be provided in a dataframe.

## 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')

2.1.2 Test for gene expression enrichment for genomic regions

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/. Alternatively also custom gene-coordinates can be provided in a dataframe.

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:10143            Pir_piriform cortex, left             3         0        0
## 2            5  Allen:10145        OlfT_Olfactory Tubercle, Left             3         0        0
## 3            5  Allen:10147        CGPo_central grey of the Pons             3         0        0
## 4            5  Allen:10148  CGPo_central grey of the pons, Left             3         0        0
## 5            5  Allen:10149 CGPo_central grey of the pons, Right             3         0        0
## 6            5  Allen:10151    LTu_Lateral Tuberal Nucleus, Left             3         0        0
##     equivalent_structures FWERs
## 1 Allen:10143;Allen:10142 0;0;0
## 2             Allen:10145 0;0;0
## 3             Allen:10147 0;0;0
## 4             Allen:10148 0;0;0
## 5             Allen:10149 0;0;0
## 6 Allen:12915;Allen:10151 0;0;0
## 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
## 276   CHMP4C     1
## 482    FABP4     1
## 483    FABP5     1
## 484    FABP9     1
## 485   FABP12     1
## 724    IMPA1     1
## 1047    PAG1     1
## 1111    PMP2     1
## 1340 SLC10A5     1
## 1386   SNX16     1
## 1683  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.

2.1.3 Custom gene-coordinates

Gene-coordinates are used when the FWER is corrected for gene length (gene_len=TRUE) or for spatial clustering of genes (genomic regions as input). Instead of using the integrated gene-coordinates, 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_len_cc = aba_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.

2.2 Test for gene expression enrichment using the Wilcoxon rank-sum test

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     22
## 2    APOL4      7
## 3     NGFR     30
## 4    NXPH4      6
## 5 C21orf59     10
## 6   CACNG2     41
## 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:13004 LHM_Lateral Hypothalamic Area, Mammillary Region             0 0.6000000     0.33
## 2            5   Allen:9455                       MBRa_Midbrain Raphe Nuclei             0 0.6000000     0.33
## 3            5   Allen:9456                 MBRa_Midbrain Raphe Nuclei, Left             0 0.6000000     0.33
## 4            5  Allen:12907                            SO_Supraoptic Nucleus             0 0.6533333     0.33
## 5            5   Allen:4593                      SO_Supraoptic Nucleus, Left             0 0.6533333     0.33
## 6            5   Allen:4633                              TubR_Tuberal Region             0 0.6533333     0.33
##   equivalent_structures          FWERs
## 1           Allen:13004 0.63;0.84;0.33
## 2            Allen:9455 0.63;0.84;0.33
## 3            Allen:9456 0.63;0.84;0.33
## 4           Allen:12907    0.63;1;0.33
## 5            Allen:4593    0.63;1;0.33
## 6            Allen:4633    0.63;1;0.33

2.3 Test for gene expression 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 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       30        9
## 2      NTS       20       14
## 3     LIPE       21        8
## 4   GALNT6       29        5
## 5      GSN       28       11
## 6   BTBD16       25       13

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 min_FWER
## 1            5   Allen:9222                    cc_corpus callosum             1     0.220     0.03
## 2            5   Allen:9241             cgb_cingulum bundle, Left             1     0.220     0.03
## 3            5   Allen:4504                   VT_Ventral Thalamus             1     0.385     0.03
## 4            5   Allen:4505             VT_Ventral Thalamus, Left             1     0.385     0.03
## 5            5   Allen:4506 R_Reticular Nucleus of Thalamus, Left             1     0.385     0.03
## 6            5   Allen:4507                 ZI_Zona Incerta, Left             1     0.385     0.03
##   equivalent_structures     FWERs
## 1 Allen:9222;Allen:9220 0.41;0.03
## 2            Allen:9241 0.41;0.03
## 3            Allen:4504 0.74;0.03
## 4            Allen:4505 0.74;0.03
## 5            Allen:4506 0.74;0.03
## 6            Allen:4507 0.74;0.03

2.4 Test for gene expression 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 [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           14       11            5        5
## 2      NTS            6       13            3        7
## 3     LIPE           12        6            0       14
## 4   GALNT6           11       14            1        5
## 5      GSN           15        6            1       10
## 6   BTBD16            6       14            1       15
## 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 14           5       
## synonymous     11           5
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
## 1            5   Allen:9072                          SN_Substantia Nigra             0 0.1200000     0.05
## 2            5   Allen:9076                   SN_Substantia Nigra, Right             0 0.1200000     0.05
## 3            5   Allen:9078 SNR_Substantia Nigra, pars reticulata, Right             0 0.1800000     0.05
## 4            5   Allen:9073                    SN_Substantia Nigra, Left             0 0.1866667     0.05
## 5            5   Allen:9557                       12_hypoglossal nucleus             0 0.2033333     0.05
## 6            5   Allen:4506        R_Reticular Nucleus of Thalamus, Left             0 0.2133333     0.05
##   equivalent_structures          FWERs
## 1            Allen:9072 0.05;0.15;0.16
## 2            Allen:9076 0.05;0.15;0.16
## 3            Allen:9078 0.05;0.33;0.16
## 4            Allen:9073 0.05;0.35;0.16
## 5            Allen:9557 0.05;0.34;0.22
## 6            Allen:4506 0.05;0.33;0.26

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.

2.5 Explore expression data

2.5.1 get_expression

The function get_expression enables the output of gene and brain region-specific expression data averaged across donors. Like in all functions of the ABAEnrichment package gene_ids can be Entrez-ID, Ensembl-ID or gene-symbol.

## get expression data for the top 5 regions
## and all input genes
## of the last aba_enrich analysis (res_conti)
top_regions = res_conti[[1]][1:5, 'structure_id']
gene_ids = res_conti[[2]][,1]
expr = get_expression(structure_ids=top_regions, gene_ids=gene_ids,
    dataset='adult')
expr[1:6,1:6]
##              BTBD16    CERS2    DOC2A     EGR4     ENC1      FRY
## Allen:9074 5.770092 6.781655 5.879127 3.303492 6.679470 7.806721
## Allen:9075 7.401949 8.372032 5.880600 3.020982 6.287626 6.989524
## Allen:9077 5.897090 6.935899 5.916721 3.047226 6.441208 7.712312
## Allen:9078 7.709733 8.661380 6.236265 3.789287 6.230635 7.118707
## Allen:9558 6.252495 7.842655 6.787737 3.537664 5.647812 7.757069
## Allen:9559 7.072817 8.412035 6.649519 3.972438 6.018460 7.554539

For the 5_stages dataset the output of get_expression is a list with a data frame for each developmental stage:

get_expression(structure_ids=c('Allen:10657','Allen:10208'),
    gene_ids=c('RFFL', 'NTS', 'LIPE'), dataset='5_stages')
## $age_category_1
##                 RFFL        NTS     LIPE
## Allen:10657 5.505681 1.84925525 2.758950
## Allen:10209 5.418012 0.23405090 2.538401
## Allen:10225 5.404229 0.09026714 2.539738
## 
## $age_category_2
##                 RFFL        NTS     LIPE
## Allen:10657 2.632211 0.43334100 4.053381
## Allen:10209 2.710142 0.12198175 5.201739
## Allen:10225 1.994992 0.05287325 2.872823
## 
## $age_category_3
##                 RFFL       NTS     LIPE
## Allen:10657 2.941152 0.2516670 4.388041
## Allen:10209 4.249908 0.3285813 4.744120
## Allen:10225 3.687618 0.3167612 5.431939
## 
## $age_category_4
##                 RFFL        NTS     LIPE
## Allen:10657 4.223104 0.02571725 5.451102
## Allen:10209 6.384951 0.05427633 8.639384
## Allen:10225 4.592677 0.05395775 4.264911
## 
## $age_category_5
##                 RFFL        NTS     LIPE
## Allen:10657 3.359006 0.43075483 5.336736
## Allen:10209 4.806099 0.03896167 5.093276
## Allen:10225 4.693686 0.03059517 4.333174

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.

2.5.2 plot_expression

The function plot_expression enables the visualization of expression data. It needs a matrix of expression data as input, like the one returned by get_expression:

## plot the expression data from above
plot_expression(expr, main="microarray expression data for adult brain")

For dataset='5_stages', get_expression returns a list of matrices; single elements of this list can be passed to plot_expression. In the following example res_devel, the enrichment analysis from above for 5 developmental stages, is used, and we want to visualize the expression of the candidate genes for the 5 brain regions with the lowest FWER for age category 2. We first extract the top 5 brain regions for age category 2 and the candidate genes used in that analysis. Then we obtain the expression of those genes in the 5 brain regions using get_expression, subset to age category 2 and pass the data to plot_expression:

## plot expression data for the top 5 regions
## of age-category 2 and all input genes
## of the developmental stage aba_enrich analysis above (res_devel)

## extract brain regions
devel_stats = res_devel[[1]]
devel_stats_2 = devel_stats[devel_stats$age_category==2,]
top_regions_2 = devel_stats_2[1:5,'structure_id']
## extract genes
genes = res_devel[[2]][,1]
## get expression for all 5 age categories
expr_all = get_expression(top_regions_2, genes, "5_stages")
## subset to age-cateogry 2
expr_2 = expr_all[["age_category_2"]]
## plot heatmap
plot_expression(expr_2, main="RPKM from RNA-seq for age_category_2")

Note that there are more than 5 brain regions in the plot, because for regions that were not sampled directly, the expression of all their sampled substructures is plotted.

Optionally also gene-associated variables (gene_vars) can be provided to create a colored side bar. This is useful to visualize the scores of genes used in an aba_enrich analysis, and gene_vars should be aba_enrich()[[2]], i.e. a data frame with all valid genes from that enrichment analysis together with their input scores.
For the hypergeometric test the colored side bar indicates candidate genes (red) and background genes (black) and for the Wilcoxon rank-sum test it 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).
In the following example we will plot the expression data for the top 5 brain regions from the 2x2 contingency table test above (res_conti), by first extracting the brain regions and genes, then getting the expression data, and finally plotting it together with the gene-associated variables from the 2x2 contingency table test as sidebar (((A+1)/(B+1)) / ((C+1)/(D+1))):

## plot expression data for the top 5 regions
## and all input genes
## from the 2x2-contingency table analysis (res_conti)

## extract brain regions
top_regions = res_conti[[1]][1:5, 'structure_id']
## extract genes
gene_ids = res_conti[[2]][,1]
## get expression 
expr = get_expression(structure_ids=top_regions, gene_ids=gene_ids, dataset='adult')
## plot expression with sidebar
plot_expression(expr, gene_vars=res_conti[[2]], main="heatmap with sidebar")

The plot_expression function is thought to give a quick overview of the expression pattern, optionally together with gene-associated variables. For more flexible heatmaps, e.g. heatmap.2 from the package gplots can be used (heatmap.2 is also used inside plot_expression):

## use heatmap.2 to allow adjusting other parameters like color
gplots::heatmap.2(expr, scale="none", col=gplots::greenred(100),
    main="custom heatmap", trace="none",
    keysize=1.4, key.xlab="expression",
    dendrogram="row", margins=c(6, 8))

2.5.3 get_name, get_sampled_substructures and get_superstructures

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.

2.5.4 get_id

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

2.5.5 get_annotated_genes

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            5  Allen:10333                            STR_striatum             0 0.3800000     0.08
## 2            2  Allen:10333                            STR_striatum             0 0.6100000     0.21
## 3            1  Allen:10294 HIP_hippocampus (hippocampal formation)             0 0.4666667     0.25
## 4            4  Allen:10294 HIP_hippocampus (hippocampal formation)             0 0.7366667     0.25
## 5            3  Allen:10333                            STR_striatum             0 0.7000000     0.27
## 6            3  Allen:10158                       Tel_telencephalon             0 0.6633333     0.32
##                 equivalent_structures          FWERs
## 1             Allen:10333;Allen:10332 0.08;0.24;0.82
## 2             Allen:10333;Allen:10332 0.21;0.87;0.75
## 3 Allen:10294;Allen:10293;Allen:10292 0.72;0.43;0.25
## 4 Allen:10294;Allen:10293;Allen:10292    0.25;0.96;1
## 5             Allen:10333;Allen:10332    0.27;0.83;1
## 6                         Allen:10158    0.32;0.67;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

2.6 Developmental effect score

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'])

3 Schematics

3.1 Schematic 1: Hypergeometric test and FWER calculation

FWER calculation

FWER calculation

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.

3.2 Schematic 2: circ_chrom option for genomic regions input

options 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. 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.

4 Session Info

sessionInfo()
## R version 3.5.1 Patched (2018-07-12 r74967)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.5 LTS
## 
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.8-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.8-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_1.12.0 BiocStyle_2.10.0    
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_0.12.19           ABAData_1.11.0         XVector_0.22.0         GenomeInfoDb_1.18.0   
##  [5] compiler_3.5.1         BiocManager_1.30.3     zlibbioc_1.28.0        bitops_1.0-6          
##  [9] tools_3.5.1            vioplot_0.2            digest_0.6.18          bit_1.1-14            
## [13] evaluate_0.12          RSQLite_2.1.1          memoise_1.1.0          DBI_1.0.0             
## [17] yaml_2.2.0             parallel_3.5.1         xfun_0.4               GenomeInfoDbData_1.2.0
## [21] GOfuncR_1.2.0          stringr_1.3.1          knitr_1.20             S4Vectors_0.20.0      
## [25] mapplots_1.5.1         gtools_3.8.1           caTools_1.17.1.1       IRanges_2.16.0        
## [29] stats4_3.5.1           rprojroot_1.3-2        bit64_0.9-7            data.table_1.11.8     
## [33] Biobase_2.42.0         AnnotationDbi_1.44.0   rmarkdown_1.10         bookdown_0.7          
## [37] gdata_2.18.0           blob_1.1.1             magrittr_1.5           GenomicRanges_1.34.0  
## [41] backports_1.1.2        gplots_3.0.1           htmltools_0.3.6        BiocGenerics_0.28.0   
## [45] KernSmooth_2.23-15     stringi_1.2.4          RCurl_1.95-4.11

5 References

[1] Hawrylycz, M.J. et al. (2012) An anatomically comprehensive atlas of the adult human brain transcriptome, Nature 489: 391-399. [https://doi.org/10.1038/nature11405]

[2] Miller, J.A. et al. (2014) Transcriptional landscape of the prenatal human brain, Nature 508: 199-206. [https://doi.org/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. [https://doi.org/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. [https://doi.org/10.1038/351652a0]