Differential expression (DE) is a common analysis that determines if the mean gene expression level differ between two groups. Most often we have two groups, normal and disease. But, in the case of TCGA, normal tissue samples are not always available. Nonetheless, samples can be grouped in many different ways, and with the number of samples and the types of data, one can be quite creative with analysis. In this work, I will be creating groups of samples based on somatic mutations, changes in the DNA of tumors, and associating it with differentially expressed genes.
The first question: "what types of mutation data are available?"
library(dplyr)
library(bigrquery)
library(scales)
library(ggplot2)
library(ISBCGCExamples)
q <- "SELECT
Variant_Classification, COUNT(*) AS n
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Study = 'BRCA'
GROUP BY
Variant_Classification"
result <- query_exec(q, project)
result## Variant_Classification n
## 1 De_novo_Start_OutOfFrame 26
## 2 Missense_Mutation 75429
## 3 Frame_Shift_Del 4056
## 4 Start_Codon_Ins 13
## 5 Stop_Codon_Ins 4
## 6 Silent 27612
## 7 5'UTR 1392
## 8 Frame_Shift_Ins 3034
## 9 In_Frame_Ins 434
## 10 Stop_Codon_Del 16
## 11 IGR 1641
## 12 Intron 6617
## 13 lincRNA 661
## 14 Nonsense_Mutation 6038
## 15 In_Frame_Del 1337
## 16 De_novo_Start_InFrame 7
## 17 RNA 2644
## 18 Splice_Site 4158
## 19 Start_Codon_SNP 97
## 20 3'UTR 2027
## 21 Nonstop_Mutation 94
## 22 Start_Codon_Del 25
This query is selecting all the variant classification labels for samples in the BRCA study using the Somatic Mutation table. The GROUP BY portion of the query ensures we only get each category once, similar to the DISTINCT keyword.
Oftentimes, people are interested in specific types of variants, such as exonic variants. Here, I'm interested in whether any type of variant might be associated with differential expression. In order to perform such an analysis, we need to define two groups. For now, let's focus on the IL10 gene, an anti-inflammatory cytokine (signalling molecule) used by the immune system. So let's find out how many individuals have a mutation in IL10, in the BRCA study.
q <- "SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'IL10'
AND Study = 'BRCA'
GROUP BY
ParticipantBarcode"
query_exec(q, project)## [1] ParticipantBarcode
## <0 rows> (or 0-length row.names)
It turns out there's no data for that gene in the BRCA study. Maybe we should find out What genes do have somatic mutations in BRCA, and how many samples? That way we can make a more educated choice.
q <- "SELECT
COUNT (DISTINCT(pb)) AS sampleN,
hugosymbols AS gene
FROM (
SELECT
ParticipantBarcode AS pb,
Hugo_Symbol AS hugosymbols
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Study = 'BRCA'
GROUP BY
hugosymbols,
pb)
GROUP BY
gene
ORDER BY
sampleN DESC"
genes <- query_exec(q, project)
head(genes)## sampleN gene
## 1 627 Unknown
## 2 335 PIK3CA
## 3 334 TP53
## 4 206 TTN
## 5 123 CDH1
## 6 117 GATA3
In this query, first on the inside, we are selecting ParticipantBarcodes and gene symbols, and grouping on those variables. This lets us get one count per gene per sample. Then from that table generated on the interior of the query, we count the number of gene symbols, giving us the number of samples with variants in a particular gene.
OK, so we see that a fair number of variants are not associated with any gene, and PIK3CA ranks first for the number of samples with mutations in that gene. Let's check our results on one gene.
q <- "SELECT count(hugosymbols)
FROM (
SELECT
ParticipantBarcode, Hugo_Symbol as hugosymbols
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'GATA3'
and Study = 'BRCA'
GROUP BY
ParticipantBarcode,
hugosymbols )"
query_exec(q, project)## f0_
## 1 117
The inner select statement produces a table with a row for each participant. Then we can count over that table to get the number of participants. This number matches what we found above. You can see that if the aggregation variable is not named, it gets a name like "f0_".
Now we need to define our groups, those with a variant in a particular gene, and those without. Let's build a table of participant barcodes with a variant in gene GATA3.
First let's get all ParticipantBarcode associated with BRCA.
q <- "
SELECT
ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Study = 'BRCA'
GROUP BY
ParticipantBarcode
"
barcodesBRCA <- query_exec(q, project)
head(barcodesBRCA)## ParticipantBarcode
## 1 TCGA-A8-A07R
## 2 TCGA-BH-A1F8
## 3 TCGA-AO-A0J8
## 4 TCGA-AN-A0XN
## 5 TCGA-A8-A097
## 6 TCGA-C8-A27B
Then, let's get the barcodes for samples with a mutation in GATA3, since it had a good number of samples according to the gene table we made above (and I'm interested in immune related genes).
q <- "
SELECT
ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'GATA3'
AND Study = 'BRCA'
GROUP BY
ParticipantBarcode
"
barcodesWithMutations <- query_exec(q, project)
head(barcodesWithMutations)## ParticipantBarcode
## 1 TCGA-OL-A5D8
## 2 TCGA-D8-A1J8
## 3 TCGA-A2-A0CV
## 4 TCGA-C8-A26V
## 5 TCGA-E2-A15J
## 6 TCGA-B6-A40C
Next, we get the participant barcodes that do not have mutations in GATA3, using the following query.
q <- "
SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE ParticipantBarcode NOT IN (
SELECT
ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'GATA3'
and Study = 'BRCA'
GROUP BY
ParticipantBarcode)
and Study = 'BRCA'
GROUP BY ParticipantBarcode"
barcodesWithOUTMutations <- query_exec(q, project)
sum(barcodesWithMutations$ParticipantBarcode %in% barcodesWithOUTMutations$ParticipantBarcode)## [1] 0
This gives us 117 samples with a variant in GATA3, 873 samples without a variant in GATA3, which matches the number of samples that have mutation data in the MAF table (990). Let's take a look at what kind of variants are found in GATA3, in case we would like to eliminate some.
q <- "
SELECT
vc,
count(vc) as num_variants_in_class
FROM (
SELECT
ParticipantBarcode,
Variant_Classification as vc
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'GATA3'
AND Study = 'BRCA'
GROUP BY
ParticipantBarcode,
vc)
GROUP BY vc
"
query_exec(q, project)## vc num_variants_in_class
## 1 Frame_Shift_Ins 62
## 2 Nonsense_Mutation 2
## 3 Splice_Site 24
## 4 Missense_Mutation 10
## 5 Stop_Codon_Ins 1
## 6 Frame_Shift_Del 19
## 7 Silent 2
So we find that frame shift insertions are the most common type of variant, followed by splice site variants.
Next, we compute the average gene expression over our defined cohorts. To do this, we first select the participants with mutations in GATA3, then in the "exterior" of the query, we select the normalized_count, which is the gene expression level, for gene GATA3, only for tumor samples (SampleTypeLetterCode = 'TP'), and only for participants with mutations in GATA3, as already defined. With the normalized_count, we can use aggregation functions to compute the mean and standard deviation on the LOG2 of expression.
q <- "
SELECT
AVG(LOG2(normalized_count+1)) as mean_expr,
STDDEV(LOG2(normalized_count+1)) as sd_expr
FROM
[isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE
HGNC_gene_symbol = 'GATA3'
AND SampleTypeLetterCode = 'TP'
AND ParticipantBarcode IN (
SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Hugo_Symbol = 'GATA3'
and Study = 'BRCA'
GROUP BY
ParticipantBarcode )
"
query_exec(q, project)## mean_expr sd_expr
## 1 14.11414 0.6700897
With the ability to compute means and stddevs, we can implement a
T statistic using unpooled standard errors.
Where x and y are the means,
q <- "
SELECT
p.gene AS gene,
p.study AS study,
p.x AS x,
p.sx2 AS sx2,
p.nx AS nx,
o.y AS y,
o.sy2 AS sy2,
o.ny AS ny,
(p.x-o.y) / SQRT((p.sx2/p.nx) + (o.sy2/o.ny)) AS T
FROM (
SELECT
Study AS study,
HGNC_gene_symbol AS gene,
AVG(LOG2(normalized_count+1)) AS x,
POW(STDDEV(LOG2(normalized_count+1)),2) AS sx2,
COUNT(ParticipantBarcode) AS nx
FROM
[isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE
HGNC_gene_symbol = 'GATA3'
AND SampleTypeLetterCode = 'TP'
AND Study = 'BRCA'
AND (ParticipantBarcode IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] AS m
WHERE
m.Hugo_Symbol = 'GATA3'
AND m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode )) /* end table of participant */
GROUP BY
gene,
study ) AS p
JOIN (
SELECT
Study AS study,
HGNC_gene_symbol AS gene,
AVG(LOG2(normalized_count+1)) AS y,
POW(STDDEV(LOG2(normalized_count+1)),2) AS sy2,
COUNT(ParticipantBarcode) AS ny
FROM
[isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE
Study = 'BRCA'
AND HGNC_gene_symbol = 'GATA3'
AND SampleTypeLetterCode = 'TP'
AND (ParticipantBarcode IN (
SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE
Study = 'BRCA'
AND (ParticipantBarcode NOT IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] AS m
WHERE
m.Hugo_Symbol = 'GATA3'
AND m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode )) /* end getting table of participants without mutations */
GROUP BY
ParticipantBarcode )) /* end getting table of participants */
GROUP BY
study,
gene ) AS o
ON
p.gene = o.gene
AND p.study = o.study
GROUP BY
gene,
study,
x,
sx2,
nx,
y,
sy2,
ny,
T
"
result1 <- query_exec(q, project)
result1## gene study x sx2 nx y sy2 ny T
## 1 GATA3 BRCA 14.11414 0.4490203 116 12.44092 4.574784 872 17.52361
A little about this query: We are selecting the gene expression from our two previously defined groups, which results in two tables that need to be joined. The two tables are named o and p (not great names, I know), which lets us perform aggregation functions on each table separately to get the mean gene expression in each group. Then, it's just a matter of computing the T statistic using the above formula.
In building this query, I found that one cannot give the source tables an alias. It seems to throw off the 'ParticipantBarcode IN ' statement, and give weird errors. Second, make sure the two tables we're joining have different aliases!
Finally, let's just "bang-the-hammer" and query ALL genes!
q <- "
SELECT
p.gene as gene,
p.study as study,
ABS(p.x - o.y) as mean_diff,
p.x as x,
p.sx2 as sx2,
p.nx as nx,
o.y as y,
o.sy2 as sy2,
o.ny as ny,
(p.x-o.y) / SQRT((p.sx2/p.nx) + (o.sy2/o.ny)) as T
FROM (
SELECT
Study as study,
HGNC_gene_symbol as gene,
AVG(LOG2(normalized_count+1)) as x,
POW(STDDEV(LOG2(normalized_count+1)),2) as sx2,
COUNT(ParticipantBarcode) as nx
FROM
[isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE
SampleTypeLetterCode = 'TP'
AND Study = 'BRCA'
and (ParticipantBarcode IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] as m
WHERE
m.Hugo_Symbol = 'GATA3'
AND m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode
)) /* end table of participant */
GROUP BY gene, study
) as p
JOIN (
SELECT
Study as study,
HGNC_gene_symbol as gene,
AVG(LOG2(normalized_count+1)) AS y,
POW(STDDEV(LOG2(normalized_count+1)),2) AS sy2,
COUNT(ParticipantBarcode) as ny
FROM
[isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE
SampleTypeLetterCode = 'TP'
and Study = 'BRCA'
AND (ParticipantBarcode IN (
SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE Study = 'BRCA'
and (ParticipantBarcode NOT IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] as m
WHERE
m.Hugo_Symbol = 'GATA3'
and m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode
)) /* end getting table of participants without mutations */
GROUP BY ParticipantBarcode
)) /* end getting table of participants */
GROUP BY
study, gene
) AS o
ON
p.gene = o.gene and p.study = o.study
GROUP BY gene, study, x, sx2, nx, y, sy2, ny, T, mean_diff
ORDER BY mean_diff DESC
"
system.time(result1 <- query_exec(q, project))##
Retrieving data: 2.9s
Retrieving data: 5.2s
## user system elapsed
## 1.280 0.114 9.194
compute_df <- function(d) {
((d$sx2/d$nx + d$sy2/d$ny)^2) /
((1/(d$nx-1))*(d$sx2/d$nx)^2 + (1/(d$ny-1))*(d$sy2/d$ny)^2)
}
result1$df <- compute_df(result1)
result1$p_value <- sapply(1:nrow(result1), function(i) 2*pt(abs(result1$T[i]), result1$df[i],lower=FALSE))
result1$fdr <- p.adjust(result1$p_value, "fdr")
result1$gene_label <- 1
result1$gene_label[result1$gene == "GATA3"] <- 2
# ordered by difference in means
head(result1)## gene study mean_diff x sx2 nx y sy2 ny
## 1 KCNJ3 BRCA 3.273704 7.485121 14.685614 116 4.211417 15.456478 872
## 2 CPB1 BRCA 3.226143 9.392237 36.573811 116 6.166094 24.273983 872
## 3 AGR3 BRCA 2.605398 11.179273 2.443981 116 8.573875 17.623450 872
## 4 TRH BRCA 2.504631 5.231328 18.403546 116 2.726697 6.268322 872
## 5 FSIP1 BRCA 2.373198 8.866036 2.245905 116 6.492839 7.712446 872
## 6 CRABP1 BRCA 2.321596 3.410742 7.469231 116 5.732338 13.186676 872
## T df p_value fdr gene_label
## 1 8.617236 149.0709 9.190821e-15 1.507825e-12 1
## 2 5.507509 136.0631 1.756605e-07 3.090353e-06 1
## 3 12.823558 393.6262 1.069316e-31 3.549239e-28 1
## 4 6.150346 125.6233 9.514207e-09 2.470345e-07 1
## 5 14.130735 237.5220 2.587392e-33 1.288198e-29 1
## 6 -8.233218 174.0920 4.162527e-14 6.050855e-12 1
# ordered by T statistic
head(result1[order(result1$T, decreasing=T), ])## gene study mean_diff x sx2 nx y sy2 ny
## 39 GATA3 BRCA 1.673222 14.114141 0.4490203 116 12.440919 4.574784 872
## 5 FSIP1 BRCA 2.373198 8.866036 2.2459046 116 6.492839 7.712446 872
## 54 TBC1D9 BRCA 1.583615 13.761795 1.0174506 116 12.178180 3.555983 872
## 77 FOXA1 BRCA 1.471983 12.926657 0.2378543 116 11.454674 8.094412 872
## 15 DNAJC12 BRCA 2.087723 10.356336 2.2985499 116 8.268613 5.297170 872
## 3 AGR3 BRCA 2.605398 11.179273 2.4439810 116 8.573875 17.623450 872
## T df p_value fdr gene_label
## 39 17.52361 513.4457 3.095801e-54 6.165288e-50 2
## 5 14.13074 237.5220 2.587392e-33 1.288198e-29 1
## 54 13.97055 239.9444 7.424160e-33 2.957043e-29 1
## 77 13.82704 947.9646 9.549048e-40 6.338976e-36 1
## 15 12.97503 193.9135 4.033130e-28 8.924421e-25 1
## 3 12.82356 393.6262 1.069316e-31 3.549239e-28 1
# ordered by FDR
head(result1[order(result1$fdr, decreasing=F), ])## gene study mean_diff x sx2 nx y sy2
## 39 GATA3 BRCA 1.673222 14.1141410 0.4490203 116 12.440919 4.574784
## 77 FOXA1 BRCA 1.471983 12.9266565 0.2378543 116 11.454674 8.094412
## 138 OPRK1 BRCA 1.283247 0.3843466 0.3377241 116 1.667593 4.605231
## 5 FSIP1 BRCA 2.373198 8.8660364 2.2459046 116 6.492839 7.712446
## 54 TBC1D9 BRCA 1.583615 13.7617954 1.0174506 116 12.178180 3.555983
## 3 AGR3 BRCA 2.605398 11.1792730 2.4439810 116 8.573875 17.623450
## ny T df p_value fdr gene_label
## 39 872 17.52361 513.4457 3.095801e-54 6.165288e-50 2
## 77 872 13.82704 947.9646 9.549048e-40 6.338976e-36 1
## 138 872 -14.17745 634.8217 7.749128e-40 6.338976e-36 1
## 5 872 14.13074 237.5220 2.587392e-33 1.288198e-29 1
## 54 872 13.97055 239.9444 7.424160e-33 2.957043e-29 1
## 3 872 12.82356 393.6262 1.069316e-31 3.549239e-28 1
qplot(data=result1, x=T, y=mean_diff, shape=as.factor(gene_label), col=as.factor(fdr < 0.05), ylab="Difference in Means", xlab="T statistic")## Warning: Removed 267 rows containing missing values (geom_point).
Wow! So did we do that right? Let's check on a single gene. We'll pull down the actual expression values, and use the R T-test.
q <- "
SELECT HGNC_gene_symbol, ParticipantBarcode, LOG2(normalized_count+1)
FROM [isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE Study = 'BRCA'
and HGNC_gene_symbol = 'GATA3'
and SampleTypeLetterCode = 'TP'
and (ParticipantBarcode IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] as m
WHERE
m.Hugo_Symbol = 'GATA3'
AND m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode
))
"
mutExpr <- query_exec(q, project) # SOME DUPLCIATES
q <- "
SELECT HGNC_gene_symbol, ParticipantBarcode, LOG2(normalized_count+1)
FROM [isb-cgc:tcga_201510_alpha.mRNA_UNC_HiSeq_RSEM]
WHERE Study = 'BRCA'
and HGNC_gene_symbol = 'GATA3'
and SampleTypeLetterCode = 'TP'
and (ParticipantBarcode IN (
SELECT
ParticipantBarcode
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls]
WHERE Study = 'BRCA'
and (ParticipantBarcode NOT IN (
SELECT
m.ParticipantBarcode,
FROM
[isb-cgc:tcga_201510_alpha.Somatic_Mutation_calls] as m
WHERE
m.Hugo_Symbol = 'GATA3'
and m.Study = 'BRCA'
GROUP BY
m.ParticipantBarcode
)) /* end getting table of participants without mutations */
GROUP BY ParticipantBarcode
)) /* end getting table of participants */
"
wtExpr <- query_exec(q, project) # SOME DUPLCIATES
t.test(mutExpr$f0_, wtExpr$f0_)##
## Welch Two Sample t-test
##
## data: mutExpr$f0_ and wtExpr$f0_
## t = 17.524, df = 513.45, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 1.485635 1.860810
## sample estimates:
## mean of x mean of y
## 14.11414 12.44092
boxplot(list(Mutation_In_GATA3=mutExpr$f0_, No_Mutation_In_GATA3=wtExpr$f0_), ylab="GATA3 LOG2 expression")
So, we have found the same gene expression means and standard deviation, degrees of freedom, and T statistic. Looks good! GATA3 has the most signficant result, interesting (or not?) since we split our samples on mutations in GATA3!
This result has been previously seen in: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4303202/
sessionInfo()## R version 3.2.4 (2016-03-10)
## Platform: x86_64-apple-darwin13.4.0 (64-bit)
## Running under: OS X 10.11.4 (El Capitan)
##
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] ISBCGCExamples_0.1.1 ggplot2_2.1.0 scales_0.4.0
## [4] bigrquery_0.2.0 dplyr_0.4.3
##
## loaded via a namespace (and not attached):
## [1] Rcpp_0.12.4 knitr_1.12.3 magrittr_1.5 munsell_0.4.3
## [5] colorspace_1.2-6 R6_2.1.2 stringr_1.0.0 httr_1.1.0
## [9] plyr_1.8.3 tools_3.2.4 parallel_3.2.4 grid_3.2.4
## [13] gtable_0.2.0 DBI_0.3.1 digest_0.6.9 openssl_0.9.2
## [17] assertthat_0.1 formatR_1.3 curl_0.9.6 evaluate_0.8.3
## [21] mime_0.4 labeling_0.3 stringi_1.0-1 jsonlite_0.9.19
## [25] markdown_0.7.7