This document contains the differential expression results from
DESeq2 for our candidate genes. These candidates are:
AR, SRD5A2, LOC113993669, ESR1, ESR2, PGR, GNRH1, PRL, PRLR, VIP, VIPR1, VIPR2, LOC113983511, OXTR, LOC113983498, AVPR1A, AVPR1B, LOC113982601
.
LOC113993669 is CYP19A1 (Aromatase)
LOC113983511 is OXT (Mesotocin)
LOC113983498 is vasotocin-neurophysin (AVP precursor)
LOC113982601 is AVPR2.
All p-values (raw and corrected) presented in this document are derived from the outputs of DESeq2 without further modification.
The next sections are scatter plots showing the expression profiles of candidate genes in each tissue. The candidate genes were first selected from the DESeq2 results based on a raw p-value of <0.05, and then plotted for validation. We excluded genes which had low R^2, and/or highly influential observations from further analysis.
rm(list= ls()[!(ls() %in% keep)])
tissue="GON"
### Create subset for Status ####
gon_key<- subset(key_behav, Tissue=="GON")
gon_key<- droplevels(gon_key)
gon_data<- data[,colnames(data) %in% rownames(gon_key)]
#remove genes with less than 5 reads
gon_data$avg_count<- apply(gon_data, 1, mean)
gon_data<- gon_data[gon_data$avg_count>5,]
gon_data$avg_count<-NULL
#remove genes where >50% of samples have 0 gene expression
gon_data$percent_0<- apply(gon_data, 1, function(x)length(x[x==0]))
thresh<- ncol(gon_data)/2
gon_data<- gon_data[gon_data$percent_0<=thresh,]
gon_data$percent_0<-NULL
dd<- DESeqDataSetFromMatrix(countData=gon_data, colData=gon_key, design= ~ Batch)
dd<- DESeq(dd)
gd<- counts(dd,normalized=TRUE)
status_results<- gd[rownames(gd) %in% candidates2,]
status_results<- as.data.frame(t(status_results))
status_results$sampleID<- rownames(status_results)
status_results<- merge(gon_key, status_results, by="sampleID")
## data for continuous variables.
gon_norm<- norm_data[[5]]
rownames(gon_norm)<- gon_norm$X
gon_norm$X<- NULL
gon_norm<- as.data.frame(t(gon_norm))
gon_norm<- gon_norm[,colnames(gon_norm) %in% candidates2]
gon_norm$sampleID<- rownames(gon_norm)
gon_norm<- merge(gon_norm, gon_key, by="sampleID")
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIPR2 | ~ Year + Status + mean_T + Status:mean_T | 155.295 | -2.027 | 0.005 | 0.496 |
AVPR1A | ~ Year + mean_T | 188.293 | 0.793 | 0.000 | 0.014 |
PRLR | ~ Year + mean_T | 1011.597 | 0.324 | 0.020 | 0.441 |
VIPR1 | ~ Year + strength.all_study | 120.522 | 0.459 | 0.022 | 0.587 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
Batch wasn’t important for anyone!
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIPR2 | ~ Status + mean_T + Status:mean_T | 155.295 | -2.019 | 0.001 | 0.379 |
VIPR2 | ~ Year + Status + mean_T + Status:mean_T | 155.295 | -2.027 | 0.005 | 0.496 |
Therefore not only is the Year variable not important for expression in any of our candidate genes, but status specific expression of VIPR2 is not sensitive to inclusion of this parameter in the model. Therefore I am going to use the model without YEAR included.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIPR2 ~ 1 | 149.6635 | NA | NA | NA |
VIPR2 ~ Status | 151.5783 | -1.9147977 | 0.7949428 | 0.9993856 |
VIPR2 ~ mean T | 149.9019 | -0.2383861 | 0.2371554 | 0.0615867 |
VIPR2 ~ Status + mean_T + Status:mean_T | 148.0384 | 1.6251157 | 0.0576544 | 0.0014815 |
Not going to accept VIPR2 because it’s not better than the null.
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
Hereafter, I exclude repeated code from presentation, but it is available on the github repository. Any variation on the code presented previously is shown.
This one is a little more complicated because PFT3 was excluded from the overall strength DE analysis owing to some genes with strong outlier effects that I couldn’t remove otherwise. So we’re also going to compare results with and without PFT3.
FALSE [1] TRUE
gene | design | baseMean | log2FoldChange | pvalue | padj | woPFT3_baseMean | woPFT3_log2FoldChange | woPFT3_pvalue | woPFT3_padj |
---|---|---|---|---|---|---|---|---|---|
AVP | ~ Batch + Status + mean_T + Status:mean_T | 123.156 | -12.764 | 0.018 | 0.997 | NA | NA | NA | NA |
AVPR1A | ~ Batch + Status + mean_T + Status:mean_T | 47.189 | 2.053 | 0.030 | 1.000 | NA | NA | NA | NA |
AVP | ~ Batch + strength.all_study | 123.156 | 2.630 | 0.008 | 0.160 | 118.037 | 4.366 | 0.000 | 0.074 |
AVP | ~ Batch + mean_T | 123.156 | 6.417 | 0.000 | 0.023 | NA | NA | NA | NA |
AVPR1B | ~ Batch + strength.all_study | 1732.190 | 0.188 | 0.081 | 0.447 | 1814.603 | 0.298 | 0.030 | 0.446 |
ESR1 | ~ Batch + Status | 1725.149 | -0.661 | 0.009 | 0.433 | NA | NA | NA | NA |
ESR2 | ~ Batch + strength.all_study | 14.951 | -0.591 | 0.038 | 0.336 | 15.496 | -0.735 | 0.043 | 0.509 |
OXT | ~ Batch + strength.all_study | 107.249 | 2.121 | 0.021 | 0.259 | 62.816 | 3.493 | 0.004 | 0.211 |
OXT | ~ Batch + mean_T | 107.249 | 4.522 | 0.001 | 0.190 | NA | NA | NA | NA |
PRL | ~ Batch + strength.all_study | 370352.282 | -0.431 | 0.000 | 0.016 | 379818.057 | -0.553 | 0.000 | 0.031 |
VIPR1 | ~ Batch + strength.all_study | 918.679 | -0.504 | 0.002 | 0.083 | 974.371 | -0.321 | 0.151 | 0.682 |
VIPR2 | ~ Batch + strength.all_study | 322.934 | -0.394 | 0.001 | 0.058 | 335.526 | -0.426 | 0.008 | 0.279 |
VIPR2 | ~ Batch + mean_T | 322.934 | -0.519 | 0.017 | 0.632 | NA | NA | NA | NA |
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AR | ~ Batch + Status + mean_T + Status:mean_T | 112.321 | -1.095 | 0.039 | 0.144 |
ESR1 | ~ Batch + Status + mean_T + Status:mean_T | 1725.149 | 0.545 | 0.014 | 0.077 |
AVPR1B | ~ Batch + Status + mean_T + Status:mean_T | 1732.190 | -0.774 | 0.000 | 0.000 |
PRL | ~ Batch + Status + mean_T + Status:mean_T | 370352.282 | 1.018 | 0.022 | 0.102 |
OXT | ~ Batch + Status + mean_T + Status:mean_T | 107.249 | -4.846 | 0.038 | 0.143 |
AVP | ~ Batch + Status + mean_T + Status:mean_T | 123.156 | -12.764 | 0.027 | 0.115 |
Batch is important for AVP.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AVP | ~ Status + mean_T + Status:mean_T | 123.156 | -11.531 | 0.028 | 0.924 |
AVP | ~ Batch + Status + mean_T + Status:mean_T | 123.156 | -12.764 | 0.018 | 0.997 |
AVPR1A | ~ Batch + Status + mean_T + Status:mean_T | 47.189 | 2.053 | 0.030 | 1.000 |
Looks like AVPR1A is not stable to model form, and while AVP is stable to model form, the gene plots for this gene are rubbish (see below). I’m still going to plot it but likely going to exclude it.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AVPR1A ~ Batch | 115.1184 | NA | 0.9089612 | NA |
AVPR1A ~ Batch + Status | 117.0048 | -1.8863751 | 0.7764506 | 0.7861074 |
AVPR1A ~ Batch + mean T | 115.9911 | -0.8727381 | 0.3710296 | 0.5124718 |
AVPR1A ~ Batch + Status * mean T | 113.1438 | 1.9745992 | 0.0726925 | 0.0300191 |
Not going to accept AVPR1A, doesn’t beat AIC test and is sensitive to inclusion of Batch.
VIPR2 was found to be associated with both mean_T and strength.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIPR2 ~ 1 | 150.6496 | NA | NA | NA |
*VIPR2 ~ Strength | 144.7318 | 5.917786 | 0.0121093 | 0.0013403 |
VIPR2 ~ mean T | 147.0481 | 3.601481 | 0.0349093 | 0.0172009 |
I’m excluding AVP and OXT owing to errant expression patterns, I’m excluding ESR1 because of an influential observation, VIPR1 and AVPR1B because results don’t hold after removal of PFT3, and AVPR1A because it failed the AIC and model parameter tests.
This tissue used additional samples for the Status comparison - thus the code used to plot these figures is presented.
rm(list= ls()[!(ls() %in% keep)])
tissue="VMH"
vmh_key<- subset(key_behav, Tissue=="VMH")
vmh_key<- droplevels(vmh_key)
vmh_data<- data[,colnames(data) %in% rownames(vmh_key)]
#remove genes with less than 5 reads
vmh_data$avg_count<- apply(vmh_data, 1, mean)
vmh_data<- vmh_data[vmh_data$avg_count>5,]
vmh_data$avg_count<-NULL
#remove genes where >50% of samples have 0 gene expression
vmh_data$percent_0<- apply(vmh_data, 1, function(x)length(x[x==0]))
thresh<- ncol(vmh_data)/2
vmh_data<- vmh_data[vmh_data$percent_0<=thresh,]
vmh_data$percent_0<-NULL
dd<- DESeqDataSetFromMatrix(countData=vmh_data, colData=vmh_key, design= ~ Batch)
dd<- DESeq(dd)
vmh_norm<- norm_data[[12]]
rownames(vmh_norm)<- vmh_norm$X
vmh_norm$X<- NULL
vmh_norm<- as.data.frame(t(vmh_norm))
vmh_norm<- vmh_norm[,colnames(vmh_norm) %in% candidates2]
vmh_norm$sampleID<- rownames(vmh_norm)
vmh_norm<- merge(vmh_norm, vmh_key, by="sampleID")
gd<- counts(dd,normalized=TRUE)
status_results<- gd[rownames(gd) %in% candidates2,]
status_results<- as.data.frame(t(status_results))
status_results$sampleID<- rownames(status_results)
status_results<- merge(vmh_key, status_results, by="sampleID")
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
GNRH1 | ~ Batch + Year + Status + mean_T + Status:mean_T | 40.945 | 7.854 | 0.000 | 0.064 |
AVPR1A | ~ Batch + Year + Status + mean_T + Status:mean_T | 131.759 | 2.227 | 0.003 | 0.455 |
PGR | ~ Batch + Year + Status + mean_T + Status:mean_T | 751.303 | -1.048 | 0.005 | 0.608 |
AR | ~ Batch + Year + strength.all_study | 951.594 | 0.632 | 0.010 | 0.104 |
AVP | ~ Batch + Year + mean_T | 1481.917 | 2.511 | 0.002 | 0.327 |
AVPR2 | ~ Batch + Year + mean_T | 10.815 | -1.125 | 0.008 | 0.588 |
ESR2 | ~ Batch + Status | 944.935 | -0.445 | 0.003 | 0.660 |
GNRH1 | ~ Batch + Year + strength.all_study | 40.945 | 2.786 | 0.034 | 0.281 |
OXT | ~ Batch + Year + mean_T | 1460.675 | 3.140 | 0.004 | 0.394 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
GNRH1 | ~ Batch + Year + Status + mean_T + Status:mean_T | 40.945 | 7.854 | 0.000 | 0.001 |
AVPR1A | ~ Batch + Year + Status + mean_T + Status:mean_T | 131.759 | 2.227 | 0.001 | 0.061 |
AR | ~ Batch + Year + Status + mean_T + Status:mean_T | 951.594 | -0.783 | 0.002 | 0.090 |
CYP19A1 | ~ Batch + Year + Status + mean_T + Status:mean_T | 5683.253 | -1.395 | 0.013 | 0.254 |
VIPR1 | ~ Batch + Year + Status + mean_T + Status:mean_T | 1186.689 | -0.522 | 0.014 | 0.258 |
PGR | ~ Batch + Year + Status + mean_T + Status:mean_T | 751.303 | -1.048 | 0.000 | 0.005 |
ESR2 | ~ Batch + Year + Status + mean_T + Status:mean_T | 820.721 | 0.026 | 0.023 | 0.329 |
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
GNRH1 | ~ Batch + Year + Status + mean_T + Status:mean_T | 40.945 | 7.854 | 0.000 | 0.064 |
AVPR1A | ~ Batch + Year + Status + mean_T + Status:mean_T | 131.759 | 2.227 | 0.003 | 0.455 |
PGR | ~ Batch + Year + Status + mean_T + Status:mean_T | 751.303 | -1.048 | 0.005 | 0.608 |
NA | NA | NA | NA | NA | NA |
Batch + Year seems to be important
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
GNRH1 ~ 1 | 141.5179 | NA | NA | NA |
GNRH1 ~ Batch + Year + Status | 142.6422 | -1.124297 | 0.0745542 | 0.8201317 |
GNRH1 ~ Batch + Year + mean T | 137.9502 | 3.567625 | 0.0373972 | NA |
GNRH1 ~ Batch + Year + Status + mean T + Status:meanT | 136.7399 | 4.777916 | 0.3478811 | 0.0000989 |
Rubbish
Rubbish plot, and requires batch + year parameters to be significant. OVERPARAMETERIZATION.
Rubbish plot, and requires batch + year parameters to be significant. OVERPARAMETERIZATION.
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
I am excluding AR on account of low Rsq,GNRH1, AVP and OXT on account of influential observations and overparameterization.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
GNRH1 | ~ Status | 5.334 | 1.749 | 0.04 | 0.581 |
In the latest revision, I realized there was a mistake in the code for AH so the results are slightly different for this nucleus and then this means we have a significant gene! But still no interaction effects.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AR | ~ Status + mean_T + Status:mean_T | 543.241 | -0.724 | 0.007 | 0.842 |
OXT | ~ Status + mean_T + Status:mean_T | 32779.237 | 1.204 | 0.033 | 0.979 |
AR | ~ mean_T | 543.241 | 0.353 | 0.016 | 0.552 |
AR | ~ strength.all_study | 543.241 | 0.192 | 0.035 | 0.584 |
AVP | ~ strength.all_study | 30862.339 | -0.504 | 0.027 | 0.581 |
AVPR2 | ~ strength.all_study | 14.871 | 0.498 | 0.024 | 0.581 |
ESR2 | ~ Status | 818.718 | -0.641 | 0.006 | 0.177 |
OXT | ~ strength.all_study | 32779.237 | -0.331 | 0.036 | 0.594 |
VIPR1 | ~ Status | 380.233 | 0.307 | 0.024 | 0.358 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
There were no relevant batch variables for PVN.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AR | ~ Status + mean_T + Status:mean_T | 543.241 | -0.724 | 0.007 | 0.842 |
OXT | ~ Status + mean_T + Status:mean_T | 32779.237 | 1.204 | 0.033 | 0.979 |
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AR ~ 1 | 157.9406 | NA | NA | NA |
AR ~ Status | 159.9100 | -1.969347 | 0.8761214 | 0.8465120 |
AR ~ mean T | 156.5464 | 1.394159 | 0.1007234 | 0.0157966 |
*AR ~ Status + mean T + Status:meanT | 152.1256 | 5.815050 | 0.0391226 | 0.0071692 |
AR ~ Strength | 155.4736 | 2.467020 | 0.0596537 | 0.0345817 |
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
OXT ~ 1 | 263.5999 | NA | NA | NA |
OXT ~ Status | 264.9400 | -1.3400716 | 0.4694425 | 0.4351629 |
OXT ~ mean T | 265.5424 | -1.9425292 | 0.8309753 | 0.8499582 |
OXT ~ Status + mean T + Status:meanT | 264.4340 | -0.8340714 | 0.1127561 | 0.0328714 |
OXT ~ Strength | 262.2297 | 1.3701415 | 0.1019376 | 0.0360569 |
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
I am excluding OXT, AVP,
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIP | ~ Status + mean_T + Status:mean_T | 3497.113 | 0.624 | 0.002 | 0.348 |
AR | ~ mean_T | 794.629 | 0.665 | 0.000 | 0.000 |
AR | ~ Batch + Status | 828.546 | 0.667 | 0.000 | 0.347 |
CYP19A1 | ~ mean_T | 1533.720 | 0.546 | 0.016 | 0.370 |
GNRH1 | ~ mean_T | 1318.753 | 0.399 | 0.006 | 0.208 |
PGR | ~ Batch + Status | 1060.152 | 0.443 | 0.006 | 0.701 |
PRLR | ~ mean_T | 4007.589 | 0.254 | 0.001 | 0.054 |
PRLR | ~ Batch + Status | 4693.061 | 0.232 | 0.016 | 0.910 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
Batch wasn’t important for anyone!
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIP | ~ Status + mean_T + Status:mean_T | 3497.113 | 0.624 | 0.002 | 0.348 |
VIP still looks good in the reduced model.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIP ~ 1 | 175.5976 | NA | NA | NA |
VIP ~ Status | 177.5578 | -1.960239 | 0.8607376 | 0.6559350 |
VIP ~ mean T | 177.5926 | -1.995031 | 0.9505483 | 0.9479884 |
VIP ~ Status + mean T + Status:mean T | 173.8064 | 1.791169 | 0.0321337 | 0.0024520 |
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AR ~ 1 | 157.3320 | NA | NA | NA |
AR ~ Status | 155.3120 | 2.019918 | 0.0774724 | 0.0004304 |
*AR ~ mean T | 144.9675 | 12.364422 | 0.0008233 | 0.0000002 |
AR ~ Status * mean T | 147.2258 | 10.106137 | 0.4983141 | 0.9714454 |
picking mean T
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
PRLR ~ 1 | 175.7063 | NA | NA | NA |
PRLR ~ Status | 175.7720 | -0.065642 | 0.2208905 | 0.0158496 |
*PRLR ~ mean T | 168.4998 | 7.206540 | 0.0074753 | 0.0007807 |
PRLR ~ Status * mean T | 169.4355 | 6.270816 | 0.1794988 | 0.1864921 |
picking meanT
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above. I’m dropping VIP, AR (Status) + PRLR (Status)
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
OXTR | ~ Batch + Year + Status + mean_T + Status:mean_T | 23.858 | 3.536 | 0.024 | 0.000 |
ESR2 | ~ Batch + Year + Status + mean_T + Status:mean_T | 233.803 | -0.910 | 0.035 | 0.000 |
AVPR1A | ~ Batch + Year + Status + mean_T + Status:mean_T | 85.431 | 2.132 | 0.036 | 0.000 |
VIP | ~ Batch + Year + mean_T | 405.910 | -0.983 | 0.011 | 0.172 |
VIPR2 | ~ Batch + Year + strength.all_study | 150.055 | -0.351 | 0.016 | 0.195 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
PRLR | ~ Batch + Year + Status + mean_T + Status:mean_T | 1680.946 | 0.4 | 0.001 | 0.193 |
Batch wasn’t important for anyone identified as being significant for our interest variables.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AVPR1A | ~ Status + mean_T + Status:mean_T | 85.431 | 1.814 | 0.019 | 1.000 |
SRD5A2 | ~ Status + mean_T + Status:mean_T | 159.287 | -0.790 | 0.028 | 1.000 |
OXTR | ~ Status + mean_T + Status:mean_T | 23.858 | 3.535 | 0.001 | 0.638 |
AVP | ~ Status + mean_T + Status:mean_T | 44.992 | 1.757 | 0.009 | 0.861 |
OXTR | ~ Batch + Year + Status + mean_T + Status:mean_T | 23.858 | 3.536 | 0.024 | 0.000 |
ESR2 | ~ Batch + Year + Status + mean_T + Status:mean_T | 233.803 | -0.910 | 0.035 | 0.000 |
AVPR1A | ~ Batch + Year + Status + mean_T + Status:mean_T | 85.431 | 2.132 | 0.036 | 0.000 |
OXTR and AVPR1A are robust to model choice.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
OXTR ~ 1 | 109.2379 | NA | NA | NA |
OXTR ~ Status | 111.1880 | -1.9500375 | 0.8440949 | 0.8310673 |
OXTR ~ mean T | 108.5720 | 0.6658985 | 0.1506183 | 0.2864778 |
OXTR ~ Status + meanT + Status:meanT | 107.1228 | 2.1151169 | 0.0756889 | 0.0014851 |
Influential observations
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AVPR1A ~ 1 | 119.4667 | NA | NA | NA |
AVPR1A ~ Status | 121.0788 | -1.6120640 | 0.5836725 | 0.7029493 |
AVPR1A ~ mean T | 121.0556 | -1.5888843 | 0.5726409 | 0.6493596 |
AVPR1A ~ Status + meanT + Status:meanT | 119.2868 | 0.1799397 | 0.0877082 | 0.0185961 |
Reject ESR2 because it was sensitive to batch parameters. OXTR because of the influential observations and AVPR1A because it didn’t make the AIC test.
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
PGR | ~ Batch + Status + mean_T + Status:mean_T | 36.895 | 1.861 | 0.005 | 0.132 |
VIP | ~ Batch + Status + mean_T + Status:mean_T | 1047.430 | 1.588 | 0.006 | 0.143 |
PRLR | ~ Batch + Status + mean_T + Status:mean_T | 734.935 | 0.703 | 0.011 | 0.200 |
ESR1 | ~ Batch + Status + mean_T + Status:mean_T | 76.686 | 1.102 | 0.041 | 0.359 |
AR | ~ Batch + strength.all_study | 149.977 | 0.215 | 0.032 | 0.530 |
CYP19A1 | ~ Batch + mean_T | 17.704 | -1.269 | 0.002 | 0.347 |
OXT | ~ Batch + Status | 8.100 | -3.087 | 0.005 | 0.175 |
PGR | ~ Batch + mean_T | 36.895 | -0.801 | 0.014 | 0.710 |
PGR | ~ Batch + strength.all_study | 36.895 | -0.656 | 0.001 | 0.138 |
VIP | ~ Batch + mean_T | 1047.430 | -0.831 | 0.009 | 0.589 |
VIP | ~ Batch + Status | 1036.449 | -0.852 | 0.046 | 0.637 |
VIP | ~ Batch + strength.all_study | 1047.430 | -0.555 | 0.007 | 0.340 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
SRD5A2 | ~ Batch + Status + mean_T + Status:mean_T | 116.627 | -0.259 | 0.036 | 0.170 |
AR | ~ Batch + Status + mean_T + Status:mean_T | 149.977 | -0.200 | 0.027 | 0.144 |
VIPR1 | ~ Batch + Status + mean_T + Status:mean_T | 667.695 | 0.291 | 0.011 | 0.086 |
Batch isn’t important for VIP, ESR1, PGR, PRLR.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIP | ~ Status + mean_T + Status:mean_T | 1047.430 | 1.722 | 0.007 | 0.350 |
ESR1 | ~ Status + mean_T + Status:mean_T | 76.686 | 1.079 | 0.037 | 0.579 |
PGR | ~ Status + mean_T + Status:mean_T | 36.895 | 1.886 | 0.003 | 0.244 |
PRLR | ~ Status + mean_T + Status:mean_T | 734.935 | 0.629 | 0.030 | 0.538 |
PGR | ~ Batch + Status + mean_T + Status:mean_T | 36.895 | 1.861 | 0.005 | 0.132 |
VIP | ~ Batch + Status + mean_T + Status:mean_T | 1047.430 | 1.588 | 0.006 | 0.143 |
PRLR | ~ Batch + Status + mean_T + Status:mean_T | 734.935 | 0.703 | 0.011 | 0.200 |
ESR1 | ~ Batch + Status + mean_T + Status:mean_T | 76.686 | 1.102 | 0.041 | 0.359 |
NA | NA | NA | NA | NA | NA |
All are robust to batch parameter.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIP ~ 1 | 190.5764 | NA | NA | NA |
VIP ~ Status | 188.6497 | 1.926707 | 0.0774627 | 0.0460933 |
VIP ~ mean T | 186.6658 | 3.910552 | 0.0302392 | 0.0085652 |
*VIP ~ Status + mean T + Status:meanT | 179.7510 | 10.825368 | 0.0153550 | 0.0069405 |
VIP ~ Strength | 188.2206 | 2.355806 | 0.0629259 | 0.0072800 |
Keep VIP
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
PGR ~ 1 | 109.73814 | NA | NA | NA |
PGR ~ Status | 109.55316 | 0.1849763 | 0.1879611 | 0.1458070 |
PGR ~ mean T | 105.36688 | 4.3712592 | 0.0244492 | 0.0143142 |
*PGR ~ Status + mean T + Status:meanT | 93.92302 | 15.8151142 | 0.0020329 | 0.0027643 |
PGR ~ Strength | 105.76276 | 3.9753773 | 0.0293451 | 0.0007425 |
Keep PGR.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
ESR1 ~ 1 | 115.4497 | NA | NA | NA |
ESR1 ~ Status | 117.4082 | -1.9584865 | 0.8560436 | 0.8211006 |
ESR1 ~ mean T | 116.0079 | -0.5581844 | 0.2848915 | 0.4275540 |
*ESR1 ~ Status + mean T + Status:mean T | 111.5403 | 3.9094611 | 0.0280822 | 0.0371658 |
Keep ESR1
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
PRLR ~ 1 | 157.7165 | NA | NA | NA |
PRLR ~ Status | 159.2028 | -1.4862765 | 0.5233260 | 0.5416420 |
PRLR ~ mean T | 159.4392 | -1.7226358 | 0.6391128 | 0.5345264 |
PRLR ~ Status + mean T + Status:mean T | 157.0247 | 0.6918305 | 0.0709455 | 0.0300391 |
Pass all except PRLR
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
Exclude PRLR, AR strength, and PGR in mean T and strength and VIP in status, mean T and Strength, OXT due to outliers
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
PRLR | ~ Batch + Status + mean_T + Status:mean_T | 1446.966 | 0.710 | 0.014 | 0.536 |
VIP | ~ Batch + Status + mean_T + Status:mean_T | 562.184 | -0.750 | 0.032 | 0.628 |
VIPR1 | ~ Batch + Status + mean_T + Status:mean_T | 104.473 | 1.010 | 0.032 | 0.628 |
CYP19A1 | ~ Batch + Status + mean_T + Status:mean_T | 1146.073 | -0.780 | 0.048 | 0.679 |
AR | ~ Batch + Status | 604.216 | -0.420 | 0.023 | 0.521 |
AVP | ~ Batch + mean_T | 65.301 | 0.835 | 0.050 | 0.480 |
AVP | ~ Batch + Status | 67.903 | 0.831 | 0.043 | 0.592 |
AVPR1A | ~ Batch + strength.all_study | 172.360 | -0.708 | 0.043 | 0.605 |
ESR2 | ~ Batch + mean_T | 182.462 | 0.810 | 0.001 | 0.079 |
PGR | ~ Batch + strength.all_study | 27.859 | -0.353 | 0.041 | 0.595 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIPR2 | ~ Batch + Status + mean_T + Status:mean_T | 1970.939 | -0.155 | 0.040 | 0.237 |
AR | ~ Batch + Status + mean_T + Status:mean_T | 655.560 | 0.436 | 0.039 | 0.233 |
ESR1 | ~ Batch + Status + mean_T + Status:mean_T | 46.675 | 0.068 | 0.013 | NA |
ESR2 | ~ Batch + Status + mean_T + Status:mean_T | 182.462 | 0.553 | 0.023 | NA |
Batch wasn’t important for any of the significant interaction models
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
CYP19A1 | ~ Status + mean_T + Status:mean_T | 1146.073 | -0.781 | 0.045 | 0.827 |
VIP | ~ Status + mean_T + Status:mean_T | 562.184 | -0.754 | 0.032 | 0.787 |
VIPR1 | ~ Status + mean_T + Status:mean_T | 104.473 | 1.015 | 0.025 | 0.768 |
PRLR | ~ Status + mean_T + Status:mean_T | 1446.966 | 0.707 | 0.026 | 0.771 |
PRLR | ~ Batch + Status + mean_T + Status:mean_T | 1446.966 | 0.710 | 0.014 | 0.536 |
VIP | ~ Batch + Status + mean_T + Status:mean_T | 562.184 | -0.750 | 0.032 | 0.628 |
VIPR1 | ~ Batch + Status + mean_T + Status:mean_T | 104.473 | 1.010 | 0.032 | 0.628 |
CYP19A1 | ~ Batch + Status + mean_T + Status:mean_T | 1146.073 | -0.780 | 0.048 | 0.679 |
All the interaction genes were stable in the reduced model.
PRLR, VIP VIPR1, CYP19A1.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
CYP19A1 ~ 1 | 178.7782 | NA | NA | NA |
CYP19A1 ~ Status | 178.6238 | 0.1544437 | 0.1910783 | 0.5105279 |
CYP19A1 ~ mean T | 180.3132 | -1.5350499 | 0.5437386 | 0.3376376 |
CYP19A1 ~ Status + mean T + Status:mean T | 178.9365 | -0.1583199 | 0.1283528 | 0.0453733 |
Does not pass AIC test.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
PRLR ~ 1 | 182.6308 | NA | NA | NA |
PRLR ~ Status | 181.4537 | 1.177132 | 0.1122971 | 0.3475519 |
PRLR ~ mean T | 184.5552 | -1.924359 | 0.8065483 | 0.5702039 |
PRLR ~ Status + mean T + Status:mean T | 179.3094 | 3.321437 | 0.1030935 | 0.0262885 |
Reject PRLR due to not sig in lm
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIP ~ 1 | 158.3812 | NA | NA | NA |
VIP ~ Status | 159.7353 | -1.354124 | 0.4742129 | 0.2119499 |
VIP ~ mean T | 156.6485 | 1.732704 | 0.0851830 | 0.0668693 |
VIP ~ Status + mean T + Status:mean T | 156.7909 | 1.590313 | 0.1197970 | 0.0319905 |
Reject VIP
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIPR1 ~ 1 | 130.2985 | NA | NA | NA |
VIPR1 ~ Status | 132.2583 | -1.959784 | 0.8582871 | 0.3738059 |
VIPR1 ~ mean T | 128.9045 | 1.394020 | 0.1007304 | 0.1297460 |
VIPR1 ~ Status + mean T + Status:mean T | 127.3370 | 2.961535 | 0.0769308 | 0.0254472 |
Reject VIPR1.
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
Reject VIP, CYP19A1 due to not passing the AIC tests, AR & AVP due to influential observations and PGR due to low R^2.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIPR1 | ~ Status + mean_T + Status:mean_T | 71.329 | 0.997 | 0.038 | 0.828 |
AVP | ~ Status + mean_T + Status:mean_T | 136.395 | 2.063 | 0.041 | 0.834 |
PGR | ~ mean_T | 28.989 | -0.530 | 0.025 | 0.499 |
PRLR | ~ Batch + Status | 259.391 | -0.423 | 0.034 | 0.592 |
VIP | ~ mean_T | 251.548 | 0.373 | 0.019 | 0.427 |
VIP | ~ strength.all_study | 251.548 | 0.229 | 0.027 | 0.427 |
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
Batch wasn’t important for anyone!
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
VIPR1 | ~ Status + mean_T + Status:mean_T | 71.329 | 0.997 | 0.038 | 0.828 |
AVP | ~ Status + mean_T + Status:mean_T | 136.395 | 2.063 | 0.041 | 0.834 |
NA | NA | NA | NA | NA | NA |
NA | NA | NA | NA | NA | NA |
VIPR is stable to model parameterization.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIPR1 ~ 1 | 97.59544 | NA | NA | NA |
VIPR1 ~ Status | 99.28283 | -1.687393 | 0.6227596 | 0.6922684 |
VIPR1 ~ mean T | 96.28066 | 1.314780 | 0.1089373 | 0.1295430 |
VIPR1 ~ Status + meanT + Status:meanT | 92.12277 | 5.472665 | 0.0313361 | 0.0375837 |
Passed AIC, and lm, but really hinges on a single observation due to reduced sample size in Ai.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AVP ~ 1 | 129.4430 | NA | NA | NA |
AVP ~ Status | 131.2984 | -1.855359 | 0.7379153 | 0.8425880 |
AVP ~ mean T | 131.1302 | -1.687228 | 0.6226676 | 0.6442814 |
AVP ~ Status + meanT + Status:meanT | 131.9397 | -2.496703 | 0.1690636 | 0.0412750 |
FAIL
distinguish between mean_T and strength
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
VIP ~ 1 | 125.3597 | NA | NA | NA |
VIP ~ Status | 125.4163 | -0.0566066 | 0.2198144 | 0.9998643 |
?VIP ~ mean T | 121.3852 | 3.9744202 | 0.0312957 | 0.0185179 |
VIP ~ Status + meanT + Status:meanT | 123.9862 | 1.3734434 | 0.3660956 | 0.2425596 |
?VIP ~ Strength | 122.2898 | 3.0698251 | 0.0473453 | 0.0265363 |
cannot distinguish between them. keep both.
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above. Exclude VIPR1,AVP, and PRLR
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AR | ~ Status + mean_T + Status:mean_T | 357.892 | -1.083 | 0.005 | 0.279 |
GNRH1 | ~ Batch + Status | 22.089 | 1.383 | 0.044 | 0.927 |
GNRH1 | ~ strength.all_study | 23.067 | 1.126 | 0.001 | NA |
SRD5A2 | ~ Batch + Status | 83.943 | 0.670 | 0.000 | 0.365 |
VIP | ~ Batch + Status | 19.078 | -1.440 | 0.034 | 0.895 |
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
Batch wasn’t important for anyone!
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
AR | ~ Status + mean_T + Status:mean_T | 357.892 | -1.083 | 0.005 | 0.279 |
AR was stable under a 1 parameter reduced model.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
AR ~ 1 | 150.1382 | NA | NA | NA |
AR ~ Status | 151.0962 | -0.9579925 | 0.3633241 | 0.3237018 |
AR ~ mean T | 152.1373 | -1.9990700 | 0.9783384 | 0.9756612 |
AR ~ Status + meanT + Status:meanT | 148.3358 | 1.8024519 | 0.0444352 | 0.0045555 |
fail
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above.
AR failed tests above, GNRH1 and VIP have influential observations.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
ESR2 | ~ Status + mean_T + Status:mean_T | 381.353 | 1.296 | 0.006 | 0.602 |
ESR1 | ~ Status + mean_T + Status:mean_T | 731.751 | 0.862 | 0.041 | 0.880 |
AR | ~ Batch + Status | 691.964 | 0.567 | 0.001 | 0.210 |
ESR2 | ~ strength.all_study | 381.353 | -0.299 | 0.047 | 0.604 |
PGR | ~ strength.all_study | 174.573 | -0.334 | 0.001 | 0.152 |
A reviewer had a concern about overparameterizing the interaction models. Therefore, for the interaction models we ran an alternate set of models to test whether number of model parameters influenced whether we got a significant result on that gene. So we ran reduced models to test whether we still those effects in reduced models, and whether batch had a significant effect on any candidate genes. The latter will decide whether I need to also include the batch variables in follow up lms and AIC.
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
Batch wasn’t important for anyone!
gene | design | baseMean | log2FoldChange | pvalue | padj |
---|---|---|---|---|---|
ESR2 | ~ Status + mean_T + Status:mean_T | 381.353 | 1.296 | 0.006 | 0.602 |
ESR1 | ~ Status + mean_T + Status:mean_T | 731.751 | 0.862 | 0.041 | 0.880 |
Both ESR1 and ESR2 pass the model stability test.
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
ESR1 ~ 1 | 168.2636 | NA | NA | NA |
ESR1 ~ Status | 169.9795 | -1.7158867 | 0.6350640 | 0.6664170 |
ESR1 ~ mean T | 167.7585 | 0.5050749 | 0.1585796 | 0.1668334 |
ESR1 ~ Status + meanT + Status:meanT | 167.4537 | 0.8098858 | 0.1017418 | 0.0411355 |
Fail AIC test
model | AIC | deltaAIC | lm_p | DESeq2_raw_p |
---|---|---|---|---|
ESR2 ~ 1 | 158.9387 | NA | NA | NA |
ESR2 ~ Status | 160.5990 | -1.6602298 | 0.6037424 | 0.4084309 |
ESR2 ~ mean T | 158.1410 | 0.7977555 | 0.1362164 | 0.1609957 |
*ESR2 ~ Status + meanT + Status:meanT | 155.8964 | 3.0423268 | 0.0482144 | 0.0060760 |
ESR2 ~ Strength | 158.3010 | 0.6377549 | 0.1479658 | 0.0467527 |
Note these decisions are made after the plotting below, but I am including a decision element in the plotting so that’s why it’s above. Excluding strength ESR2 due to low R^2
Scaled according to individual gene’s expression.
FALSE png
FALSE 2
FALSE R version 4.0.2 (2020-06-22)
FALSE Platform: x86_64-w64-mingw32/x64 (64-bit)
FALSE Running under: Windows 10 x64 (build 19045)
FALSE
FALSE Matrix products: default
FALSE
FALSE locale:
FALSE [1] LC_COLLATE=English_Australia.1252 LC_CTYPE=English_Australia.1252
FALSE [3] LC_MONETARY=English_Australia.1252 LC_NUMERIC=C
FALSE [5] LC_TIME=English_Australia.1252
FALSE
FALSE attached base packages:
FALSE [1] grid parallel stats4 stats graphics grDevices utils
FALSE [8] datasets methods base
FALSE
FALSE other attached packages:
FALSE [1] RColorBrewer_1.1-2 MASS_7.3-51.6
FALSE [3] dplyr_1.0.7 ggnetwork_0.5.8
FALSE [5] igraph_1.2.6 ggfortify_0.4.11
FALSE [7] GGally_2.1.0 effectsize_0.8.6
FALSE [9] car_3.0-8 carData_3.0-4
FALSE [11] viridis_0.6.1 viridisLite_0.4.0
FALSE [13] rrvgo_1.0.2 UpSetR_1.4.0
FALSE [15] limma_3.44.3 flashClust_1.01-2
FALSE [17] org.Hs.eg.db_3.11.4 AnnotationDbi_1.50.0
FALSE [19] clusterProfiler_3.16.1 fdrtool_1.2.15
FALSE [21] ggpubr_0.4.0 gridExtra_2.3
FALSE [23] wesanderson_0.3.6 VennDiagram_1.6.20
FALSE [25] futile.logger_1.4.3 kableExtra_1.3.1
FALSE [27] WGCNA_1.69 fastcluster_1.1.25
FALSE [29] dynamicTreeCut_1.63-1 pheatmap_1.0.12
FALSE [31] lubridate_1.8.0 DESeq2_1.28.1
FALSE [33] SummarizedExperiment_1.18.1 DelayedArray_0.14.0
FALSE [35] matrixStats_0.56.0 Biobase_2.48.0
FALSE [37] GenomicRanges_1.40.0 GenomeInfoDb_1.24.0
FALSE [39] IRanges_2.22.2 S4Vectors_0.26.1
FALSE [41] BiocGenerics_0.34.0 stringr_1.4.0
FALSE [43] PCAtools_2.0.0 cowplot_1.0.0
FALSE [45] lattice_0.20-41 ggrepel_0.8.2
FALSE [47] ggplot2_3.4.4 reshape2_1.4.4
FALSE [49] plyr_1.8.6
FALSE
FALSE loaded via a namespace (and not attached):
FALSE [1] ragg_1.2.1 tidyr_1.1.4
FALSE [3] acepack_1.4.1 bit64_0.9-7
FALSE [5] knitr_1.29 irlba_2.3.3
FALSE [7] data.table_1.13.0 rpart_4.1-15
FALSE [9] RCurl_1.98-1.2 doParallel_1.0.15
FALSE [11] generics_0.0.2 snow_0.4-3
FALSE [13] preprocessCore_1.50.0 lambda.r_1.2.4
FALSE [15] RSQLite_2.2.0 europepmc_0.4
FALSE [17] bit_4.0.4 enrichplot_1.8.1
FALSE [19] webshot_0.5.2 xml2_1.3.2
FALSE [21] httpuv_1.5.4 assertthat_0.2.1
FALSE [23] xfun_0.29 hms_1.1.1
FALSE [25] jquerylib_0.1.4 evaluate_0.14
FALSE [27] promises_1.1.1 fansi_0.4.1
FALSE [29] progress_1.2.2 readxl_1.3.1
FALSE [31] DBI_1.1.0 geneplotter_1.66.0
FALSE [33] htmlwidgets_1.5.3 reshape_0.8.8
FALSE [35] purrr_0.3.4 ellipsis_0.3.2
FALSE [37] backports_1.1.7 insight_0.19.6
FALSE [39] annotate_1.66.0 gridBase_0.4-7
FALSE [41] vctrs_0.6.5 abind_1.4-5
FALSE [43] withr_2.5.2 ggforce_0.3.2
FALSE [45] triebeard_0.3.0 treemap_2.4-2
FALSE [47] checkmate_2.0.0 prettyunits_1.1.1
FALSE [49] cluster_2.1.0 DOSE_3.14.0
FALSE [51] crayon_1.4.2 genefilter_1.70.0
FALSE [53] pkgconfig_2.0.3 slam_0.1-47
FALSE [55] labeling_0.3 tweenr_1.0.1
FALSE [57] nlme_3.1-148 wordcloud_2.6
FALSE [59] nnet_7.3-14 rlang_1.1.2
FALSE [61] lifecycle_1.0.4 downloader_0.4
FALSE [63] rsvd_1.0.3 cellranger_1.1.0
FALSE [65] polyclip_1.10-0 datawizard_0.9.0
FALSE [67] Matrix_1.6-4 urltools_1.7.3
FALSE [69] base64enc_0.1-3 ggridges_0.5.3
FALSE [71] png_0.1-7 parameters_0.21.3
FALSE [73] bitops_1.0-6 blob_1.2.1
FALSE [75] DelayedMatrixStats_1.10.0 qvalue_2.20.0
FALSE [77] jpeg_0.1-8.1 rstatix_0.7.0
FALSE [79] gridGraphics_0.5-1 ggsignif_0.6.0
FALSE [81] scales_1.3.0 memoise_1.1.0
FALSE [83] magrittr_2.0.1 zlibbioc_1.34.0
FALSE [85] compiler_4.0.2 scatterpie_0.1.5
FALSE [87] dqrng_0.2.1 cli_3.6.1
FALSE [89] XVector_0.28.0 htmlTable_2.0.1
FALSE [91] formatR_1.8 Formula_1.2-3
FALSE [93] mgcv_1.8-31 tidyselect_1.1.0
FALSE [95] stringi_1.4.6 forcats_0.5.1
FALSE [97] textshaping_0.3.6 highr_0.8
FALSE [99] yaml_2.2.1 GOSemSim_2.14.2
FALSE [101] BiocSingular_1.4.0 locfit_1.5-9.4
FALSE [103] latticeExtra_0.6-29 fastmatch_1.1-0
FALSE [105] tools_4.0.2 rio_0.5.16
FALSE [107] rstudioapi_0.13 foreach_1.5.0
FALSE [109] foreign_0.8-80 farver_2.0.3
FALSE [111] ggraph_2.0.5 digest_0.6.25
FALSE [113] rvcheck_0.1.8 BiocManager_1.30.10
FALSE [115] shiny_1.5.0 Rcpp_1.0.7
FALSE [117] broom_0.7.10 later_1.1.0.1
FALSE [119] httr_1.4.2 colorspace_1.4-1
FALSE [121] rvest_1.0.2 XML_3.99-0.5
FALSE [123] splines_4.0.2 graphlayouts_0.7.1
FALSE [125] ggplotify_0.0.5 systemfonts_1.0.3
FALSE [127] xtable_1.8-4 jsonlite_1.7.2
FALSE [129] futile.options_1.0.1 tidygraph_1.2.0
FALSE [131] NLP_0.2-1 R6_2.4.1
FALSE [133] tm_0.7-8 Hmisc_4.4-0
FALSE [135] pillar_1.6.4 htmltools_0.5.2
FALSE [137] mime_0.9 glue_1.5.1
FALSE [139] fastmap_1.1.0 BiocParallel_1.22.0
FALSE [141] codetools_0.2-16 fgsea_1.14.0
FALSE [143] utf8_1.1.4 tibble_3.1.6
FALSE [145] curl_4.3 zip_2.0.4
FALSE [147] GO.db_3.11.4 openxlsx_4.1.5
FALSE [149] survival_3.2-3 rmarkdown_2.11
FALSE [151] munsell_0.5.0 DO.db_2.9
FALSE [153] GenomeInfoDbData_1.2.3 iterators_1.0.12
FALSE [155] impute_1.62.0 haven_2.3.1
FALSE [157] gtable_0.3.0 bayestestR_0.13.1