Help with multi-factor Deseq analysis
0
0
Entering edit mode
6.3 years ago
MAPK ★ 2.1k

Hi All, I need help to resolve this problem with deseq. I have two cultivars (R and S), two time points (5d and 30d) and four treatments (Scn, Aph, ScnAph, Ctr). There are two sets of controls (Ctr) for each cultivar and I have a total of 48 samples. I need to compare:

  1. Scn Vs Ctr within resistance within 5 day
  2. Scn Vs Aph within susceptible within 5 day
  3. Scn Vs Ctr within resistance within 30 day
  4. Scn Vs Aph within susceptible within 30 day
  5. R Vs S in whole dataset

Can someone please guide me through this. I have tried this model which looks like this, but I am not sure if this is the correct way of doing it. Thanks for your help in advance.

dds <- DESeqDataSetFromMatrix(countData=count.mat,colData=cond,design= ~cultivar+time+cultivar:time)
##you  relevel 5d so you'll do "cultivarR.time30d"
dds$time <- relevel(dds$time, ref = "5d")   
##you relevel S so you'll do "cultivarR.time5d"
dds$cultivar <- relevel(dds$cultivar, ref = "S") 
dds = DESeq(dds, test="LRT", reduced= ~cultivar + time)
#5d vs 30d within resistance
##in case where dds$time <- relevel(dds$time, ref = "S")
resultsNames(dds)
# > resultsNames(dds)
# [1] "Intercept"        "cultivar_R_vs_S"  "time_5d_vs_30d"   "cultivarR.time5d"
#5d vs 30d within resistance
tt<- results(dds, contrast=list(c("time_5d_vs_30d", "cultivarR.time5d")), test="Wald")

My data:

my count.mat looks like this:

count.mat <- structure(list(sam5_4a_counts = c(26, 28, 1, 137, 3, 0), sam5_4b_counts = c(13, 
6, 0, 95, 3, 0), sam5_4c_counts = c(1, 7, 0, 41, 2, 0), sam5_5a_counts = c(28, 
15, 0, 84, 0, 0), sam5_5b_counts = c(12, 29, 1, 97, 2, 1), sam5_5c_counts = c(10, 
11, 0, 77, 2, 0), sam5_6a_counts = c(42, 24, 0, 139, 4, 0), sam5_6b_counts = c(29, 
28, 1, 166, 1, 1), sam5_6c_counts = c(29, 46, 0, 112, 5, 3), 
    sam5_7a_counts = c(7, 7, 1, 65, 0, 0), sam5_7b_counts = c(37, 
    10, 0, 108, 4, 0), sam5_7c_counts = c(7, 4, 0, 47, 0, 0), 
    sam5_1a_counts = c(44, 56, 4, 107, 2, 0), sam5_1b_counts = c(13, 
    11, 3, 44, 1, 0), sam5_1c_counts = c(39, 55, 1, 166, 1, 0
    ), sam5_2a_counts = c(4, 8, 1, 75, 1, 0), sam5_2b_counts = c(126, 
    160, 10, 414, 5, 1), sam5_2c_counts = c(28, 37, 1, 209, 3, 
    1), sam5_3a_counts = c(38, 70, 5, 138, 3, 1), sam5_3b_counts = c(132, 
    218, 6, 390, 14, 5), sam5_3c_counts = c(10, 2, 0, 39, 0, 
    1), sam5_8a_counts = c(19, 37, 1, 140, 1, 1), sam5_8b_counts = c(5, 
    12, 0, 63, 1, 0), sam5_8c_counts = c(27, 14, 1, 90, 0, 0), 
    sam30_4a_counts = c(29, 31, 0, 68, 2, 0), sam30_4b_counts = c(32, 
    24, 0, 70, 1, 0), sam30_4c_counts = c(13, 8, 0, 38, 2, 1), 
    sam30_5a_counts = c(22, 14, 0, 104, 2, 0), sam30_5b_counts = c(37, 
    49, 2, 88, 1, 0), sam30_5c_counts = c(37, 84, 1, 106, 0, 
    1), sam30_6a_counts = c(74, 58, 3, 110, 2, 1), sam30_6b_counts = c(68, 
    183, 3, 150, 2, 1), sam30_6c_counts = c(38, 86, 1, 161, 1, 
    0), sam30_7a_counts = c(21, 27, 0, 93, 2, 0), sam30_7b_counts = c(27, 
    20, 0, 89, 0, 1), sam30_7c_counts = c(24L, 23L, 0L, 91L, 
    1L, 0L), sam30_1a_counts = c(5, 7, 0, 16, 0, 0), sam30_1b_counts = c(38, 
    35, 3, 102, 0, 0), sam30_1c_counts = c(55, 26, 0, 136, 2, 
    0), sam30_2a_counts = c(20, 6, 0, 65, 2, 0), sam30_2b_counts = c(10, 
    5, 0, 43, 0, 0), sam30_2c_counts = c(86, 88, 7, 167, 6, 0
    ), sam30_3a_counts = c(39, 19, 1, 132, 3, 0), sam30_3b_counts = c(30, 
    31, 0, 113, 2, 0), sam30_3c_counts = c(59, 35, 0, 104, 3, 
    0), sam30_8a_counts = c(37, 22, 0, 133, 1, 0), sam30_8b_counts = c(28, 
    20, 0, 150, 2, 0), sam30_8c_counts = c(13, 20, 1, 104, 0, 
    0)), row.names = c("Glyma.01G000100", "Glyma.01G000200", 
"Glyma.01G000300", "Glyma.01G000400", "Glyma.01G000500", "Glyma.01G000600"
), class = "data.frame")

my condition dataframe cond.

cond <- structure(list(cultivar = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("R", "S"), class = "factor"), 
    time = structure(c(2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
    2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 
    1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
    1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L), .Label = c("30d", "5d"), class = "factor"), 
    treatment = structure(c(3L, 3L, 3L, 1L, 1L, 1L, 4L, 4L, 4L, 
    2L, 2L, 2L, 3L, 3L, 3L, 1L, 1L, 1L, 4L, 4L, 4L, 2L, 2L, 2L, 
    3L, 3L, 3L, 1L, 1L, 1L, 4L, 4L, 4L, 2L, 2L, 2L, 3L, 3L, 3L, 
    1L, 1L, 1L, 4L, 4L, 4L, 2L, 2L, 2L), .Label = c("Aph", "Ctr", 
    "Scn", "ScnAph"), class = "factor")), row.names = c(NA, -48L
), class = "data.frame")
deseq2 rna-seq • 3.0k views
ADD COMMENT
2
Entering edit mode

I think that you should consider merging your cultivar and treatment variables:

cond$CultivarTreatment <- paste(cond$cultivar, cond$treatment, sep="_")

head(cond)
  cultivar time treatment CultivarTreatment
1        R   5d       Scn             R_Scn
2        R   5d       Scn             R_Scn
3        R   5d       Scn             R_Scn
4        R   5d       Aph             R_Aph
5        R   5d       Aph             R_Aph
6        R   5d       Aph             R_Aph

....and then using:

dds <- DESeqDataSetFromMatrix(countData = count.mat,
  colData = cond, design = ~ CultivarTreatment + time + CultivarTreatment:time)

dds <- DESeq(dds, test = 'LRT', reduced = ~ CultivarTreatment + time)

Best person to ask is, of course, Michael Love, who is more active on the Bioconductor forums. He'll respond pretty quickly there.

ADD REPLY
0
Entering edit mode

Thanks, Kevin. I will post this in bioconductor as well.

ADD REPLY
0
Entering edit mode

I think that the way I've stated will allow you to achieve numbers 1-4 in your list, but possibly not #5. Interested to see what Michael Love suggests.

If anyone else here has suggestions, please chip in.

ADD REPLY
0
Entering edit mode

@Kevin Blighe, So I got the following results after running your code. Now how do I get #1 to #4? Thanks again for your help.

> resultsNames(dds)
 [1] "Intercept"                           "CultivarTreatment_R_Ctr_vs_R_Aph"    "CultivarTreatment_R_Scn_vs_R_Aph"    "CultivarTreatment_R_ScnAph_vs_R_Aph"
 [5] "CultivarTreatment_S_Aph_vs_R_Aph"    "CultivarTreatment_S_Ctr_vs_R_Aph"    "CultivarTreatment_S_Scn_vs_R_Aph"    "CultivarTreatment_S_ScnAph_vs_R_Aph"
 [9] "time_5d_vs_30d"                      "CultivarTreatmentR_Ctr.time5d"       "CultivarTreatmentR_Scn.time5d"       "CultivarTreatmentR_ScnAph.time5d"   
[13] "CultivarTreatmentS_Aph.time5d"       "CultivarTreatmentS_Ctr.time5d"       "CultivarTreatmentS_Scn.time5d"       "CultivarTreatmentS_ScnAph.time5d"
ADD REPLY
0
Entering edit mode

I see that you got similar advice on Bioconductor. Would it not make more sense to set 5d as the reference level for time? It looks like you have 30d as the reference level. If you switch that around, then it should be (I think):

# Scn Vs Ctr within resistance within 5 day
results(dds, name="CultivarTreatment_R_Scn_vs_R_Ctr", test="Wald") # due to the fact that 5d is reference level

# Scn Vs Aph within susceptible within 5 day
results(dds, name="CultivarTreatment_S_Scn_vs_S_Aph", test="Wald")

# Scn Vs Ctr within resistance within 30 day
results(dds, contrast=list(c("CultivarTreatmentR_Scn.time30d", "CultivarTreatmentR_Ctr.time30d")), test="Wald")

Scn Vs Aph within susceptible within 30 day
results(dds, contrast=list(c("CultivarTreatmentS_Scn.time30d", "CultivarTreatmentS_Aph.time30d")), test="Wald")

I would just confirm this with Michael Love

ADD REPLY
1
Entering edit mode

Thank you so much, Kevin. That's right, releveling it to ref=5 day makes more sense.

ADD REPLY
1
Entering edit mode

I see that it was recommended to combine all 3 together with paste. I had considered that but was not entirely sure of the validity in the context of a DESeq2 design formula.

ADD REPLY
0
Entering edit mode

Yes, that's right. However, after merging all three columns (as TimeTreatmentCultivar), I tried and got the following error below. How do I reduce it properly? Thank you for your help. my code:

 cond$TimeTreatmentCultivar<- as.factor(paste(cond$cultivar,cond$time,cond$treatment, sep = "_"))
    dds <- DESeqDataSetFromMatrix(countData=count.mat,colData=cond,design= ~TimeTreatmentCultivar)
    dds = DESeq(dds, test="LRT", reduced= ~ TimeTreatmentCultivar)

Error:

estimating size factors
estimating dispersions
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing
Error in nbinomLRT(object, full = full, reduced = reduced, quiet = quiet,  : 
  less than one degree of freedom, perhaps full and reduced models are not in the correct order
ADD REPLY
0
Entering edit mode

I thought that, in this situation, you could avoid the reduced step.

Also, given the complex experimental set-up that you have, I also thought that, perhaps, you could perform everything separately for R and S, although that introduces some extra bias into the results.

ADD REPLY
0
Entering edit mode

So should I just do dds = DESeq(dds) instead of dds = DESeq(dds, test="LRT", reduced= ~ TimeTreatmentCultivar)? I also trieddds = DESeq(dds, test="LRT"), but got this error below. Or should I just do reduce = ~1?

Error in DESeq(dds, test = "LRT") : 
  likelihood ratio test requires a 'reduced' design, see ?DESeq
ADD REPLY
1
Entering edit mode

I think that you should just do DESeq(dds). When you generate your results, you don't appear to use the LRT anywhere; instead, Wald is specified. As your aim is to just compare pairs of levels, you don't require the LRT functionality.

ADD REPLY

Login before adding your answer.

Traffic: 1802 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6