Design/contrast matrix for Multivariate experimental design in limma

Question

[1234anjalianjali1234]

Hello,

I have four number of Two-color array chips (.gpr files/ genepix) with different time points, infecting organism and with dye swaps and some having biological replicate. it's getting complicated for me to make design/contrast matrix to find differential gene expression.

I have done normalization step with all these chips and also merge them using InSilicoDb package using Combat method. Could anyone help me to make design matrix for these chips in combined form?

Sample description with Cy3/Cy5 information given below.

WT = Wounded and Treated;
T = Treated;
CDLV = Carborundum-dusted leaves treated with Virus;
CDLW = Carborundum-dusted leavestreated with Water

 

Sample_title             Organism    Geo      protocol_1  Label_1  protocol_2  label_ch2
ctrl_vs_race0_Scjm_72h_1    Fungus   GSM200000  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Scjm_72h_2    Fungus   GSM200001  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Scjm_72h_3    Fungus   GSM200002  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Scjm_96h_1    Fungus   GSM200003  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Scjm_96h_2    Fungus   GSM200004  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_72h_1    Fungus   GSM200005  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_72h_2    Fungus   GSM200006  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_72h_3    Fungus   GSM200007  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_96h_1    Fungus   GSM200008  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_96h_2    Fungus   GSM200009  Treated      Cy3    Control       Cy5
ctrl_vs_race0_Stbr_96h_3    Fungus   GSM200010  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_72h_1  Fungus   GSM200011  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_72h_2  Fungus   GSM200012  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_72h_3  Fungus   GSM200013  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_96h_1  Fungus   GSM200014  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_96h_2  Fungus   GSM200015  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Scjm_96h_3  Fungus   GSM200016  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_72h_1  Fungus   GSM200017  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_72h_2  Fungus   GSM200018  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_72h_3  Fungus   GSM200019  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_96h_1  Fungus   GSM200020  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_96h_2  Fungus   GSM200021  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_96h_3  Fungus   GSM200022  Treated      Cy3    Control       Cy5
ctrl_vs_complex_Stbr_72h_1FD Fungus  GSM200023  Control      Cy3    Treated       Cy5

Sample                   Organism     Geo    protocol_1  Label_1  protocol_2 label_2
ctrl_vs_1h_infested1           Insect GSM200024  T 1hr     Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested2           Insect GSM200025  T 1hr     Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested3           Insect GSM200026  T 1 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested1swap       Insect GSM200027  T 1hr     Cy5   Control 4 hr  Cy3
ctrl_vs_1h_infested2swap       Insect GSM200028  T 1hr     Cy5   Control 4 hr  Cy3
ctrl_vs_1h_infested3swap       Insect GSM200029  T 1 hr    Cy5   Control 4 hr  Cy3
wounded_vs_wounded_spit1       Insect GSM200030  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit2       Insect GSM200031  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit3       Insect GSM200032  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit1swap   Insect GSM200033  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wounded_vs_wounded_spit2swap   Insect GSM200034  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wounded_vs_wounded_spit3swap   Insect GSM200035  WT 4 hr   Cy5   Wounded 4 hr  Cy3
ctrl_vs_4h_infested1           Insect GSM200036  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested2           Insect GSM200037  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested3           Insect GSM200038  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested1swap       Insect GSM200039  T 4 hr    Cy5   Control 4 hr  Cy3
ctrl_vs_4h_infested2swap       Insect GSM200040  T 4 hr    Cy5   Control 4 hr  Cy3
ctrl_vs_4h_infested3swap       Insect GSM200041  T 4 hr    Cy5   Control 4 hr  Cy3
wd_systemic_vs_spit_systemic1  Insect GSM200042  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic2  Insect GSM200043  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic3  Insect GSM200044  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic1s Insect GSM200045  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wd_systemic_vs_spit_systemic2s Insect GSM200046  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wd_systemic_vs_spit_systemic3s Insect GSM200047  WT 4 hr   Cy5   Wounded 4 hr  Cy3

Sample_title              Organism    Geo     protocol_1  Label_1  protocol_2   label_2
non_GM_D_vs_GM_A_Pinf_1    Fungus  GSM200048  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_D_vs_GM_A_water_1   Fungus  GSM200049  Control_24h  Cy3  Control_24h  Cy5
non_GM_E_vs_GM_B_Pinf_1    Fungus  GSM200050  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_E_vs_GM_B_water_1   Fungus  GSM200051  Control_24h  Cy3  Control_24h  Cy5
non_GM_F_vs_GM_C_Pinf_1    Fungus  GSM200052  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_F_vs_GM_C_water_1   Fungus  GSM200053  Control_24h  Cy3  Control_24h  Cy5
non_GM_D_vs_GM_A_Pinf_2    Fungus  GSM200054  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_D_vs_GM_A_water_2   Fungus  GSM200055  Control_24h  Cy3  Control_24h  Cy5
non_GM_E_vs_GM_B_Pinf_2    Fungus  GSM200056  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_E_vs_GM_B_water_2   Fungus  GSM200057  Control_24h  Cy3  Control_24h  Cy5
non_GM_F_vs_GM_C_Pinf_2    Fungus  GSM200058  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_F_vs_GM_C_water_2   Fungus  GSM200059  Control_24h  Cy3  Control_24h  Cy5
non_GM_D_vs_GM_A_Pinf_3    Fungus  GSM200060  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_D_vs_GM_A_water_3   Fungus  GSM200061  Control_24h  Cy3  Control_24h  Cy5
non_GM_E_vs_GM_B_Pinf_3    Fungus  GSM200062  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_E_vs_GM_B_water_3   Fungus  GSM200063  Control_24h  Cy3  Control_24h  Cy5
non_GM_F_vs_GM_C_Pinf_3    Fungus  GSM200064  Treated_24h  Cy3  Treated_24h  Cy5
non_GM_F_vs_GM_C_water_3   Fungus  GSM200065  Control_24h  Cy3  Control_24h  Cy5
non_GM_D_vs_GM_A_water_3FD Fungus  GSM200066  Control_24h  Cy5  Control_24h  Cy3

!Sample_title               Organism    Geo    treatment_1  Label_1 treatment_2 Label_2
mock_vs_pvy_infected_1dpi_1.1  Virus GSM200067 CDLV_1dpi    Cy3 CDLW_1dpi    Cy5
mock_vs_pvy_infected_1dpi_1.2  Virus GSM200068 CDLV_1dpi    Cy3 CDLW_1dpi    Cy5
mock_vs_pvy_infected_1dpi_2.1  Virus GSM200069 CDLV_1dpi    Cy3 CDLW_1dpi    Cy5
mock_vs_pvy_infected_1dpi_2.2  Virus GSM200070 CDLV_1dpi    Cy3 CDLW_1dpi    Cy5
mock_vs_pvy_infected_3dpi_1.1  Virus GSM200071 CDLV_3dpi    Cy3 CDLW_3dpi    Cy5
mock_vs_pvy_infected_3dpi_1.2  Virus GSM200072 CDLV_3dpi    Cy3 CDLW_3dpi    Cy5
mock_vs_pvy_infected_3dpi_1_FD Virus GSM200073 CDLV_3dpi    Cy5 CDLW_3dpi    Cy3
mock_vs_pvy_infected_3dpi_2.1  Virus GSM200074 CDLV_3dpi    Cy3 CDLW_3dpi    Cy5
mock_vs_pvy_infected_3dpi_2.2  Virus GSM200075 CDLV_3dpi    Cy3 CDLW_3dpi    Cy5
mock_vs_pvy_infected_6dpi_1.1  Virus GSM200076 CDLV_6dpi    Cy3 CDLW_6dpi    Cy5
mock_vs_pvy_infected_6dpi_1.2  Virus GSM200077 CDLV_6dpi    Cy3 CDLW_6dpi    Cy5
mock_vs_pvy_infected_6dpi_2.1  Virus GSM200078 CDLV_6dpi    Cy3 CDLW_6dpi    Cy5
mock_vs_pvy_infected_6dpi_2.2  Virus GSM200079 CDLV_6dpi    Cy3 CDLW_6dpi    Cy5
mock_vs_pvy_systemic_1dpi_1.1  Virus GSM200080 Treated_1dpi Cy3 Treated_1dpi Cy5
mock_vs_pvy_systemic_1dpi_1.2  Virus GSM200081 Treated_1dpi Cy3 Treated_1dpi Cy5
mock_vs_pvy_systemic_1dpi_2.1  Virus GSM200082 Treated_1dpi Cy3 Treated_1dpi Cy5
mock_vs_pvy_systemic_1dpi_2.2  Virus GSM200083 Treated_1dpi Cy3 Treated_1dpi Cy5
mock_vs_pvy_systemic_3dpi_1.1  Virus GSM200084 Treated_3dpi Cy3 Treated_3dpi Cy5
mock_vs_pvy_systemic_3dpi_1.2  Virus GSM200085 Treated_3dpi Cy3 Treated_3dpi Cy5
mock_vs_pvy_systemic_3dpi_1_FD Virus GSM200086 Treated_3dpi Cy5 Treated_3dpi Cy3
mock_vs_pvy_systemic_3dpi_2.1  Virus GSM200087 Treated_3dpi Cy3 Treated_3dpi Cy5
mock_vs_pvy_systemic_3dpi_2.2  Virus GSM200088 Treated_3dpi Cy3 Treated_3dpi Cy5
mock_vs_pvy_systemic_6dpi_1.1  Virus GSM200089 Treated_6dpi Cy3 Treated_6dpi Cy5
mock_vs_pvy_systemic_6dpi_1.2  Virus GSM200090 Treated_6dpi Cy3 Treated_6dpi Cy5
mock_vs_pvy_systemic_6dpi_2.1  Virus GSM200091 Treated_6dpi Cy3 Treated_6dpi Cy5
mock_vs_pvy_systemic_6dpi_2.2  Virus GSM200092 Treated_6dpi Cy3 Treated_6dpi Cy5

[Kevin Blighe]

Some things do not make sense, for now at least. Can you clarify the following:

1. Why did you use Combat? - my recommendation is to never use Combat. You likely used Combat to correct for a known batch effect, in which case you should just include 'batch' in your design model.
2. You have fungus, insect, and virus under [infecting] 'Organism' - you obviously do not want to compare across these?; or are you attempting to control for this confounding factor?

-----------------------------------------

Can you list the contrasts that you want to see? For example:

• Fungus: Treated_24h Versus Control_24h
• Inect: T 1 hr Versus WT 4 hr
• et cetera

Finally, can you take a look at my tutorial for 2-colour arrays and see if you can get any help there. Note, that, for the function read.maimages(), you would select source="genepix"

[1234anjalianjali1234]

HI kevin,

1.YES, I have used combat to remove the batch effect and want to include batch effect in design/contrast matrix.

2.Actually, I want to find DEG in response to Biotic stress on plants, so yes I want to compare across these organisms with respect to given different time points.

[Kevin Blighe]

Okay, but I and many others would never recommend Combat. By using it, you run the risk of modifying your data unexpectedly, possibly introducing even more bias into your results than there would have been had you not used Combat.

Your experimental set-up does look complex. Please note, however, that, if you have processed your data correctly, then you should have produced a MAlist for each sample. The log base 2 expression values for each sample can then be accessed via the M object. For example, MyObject$M

In your limma design model, I would include at least

~ Organism + treatment_1 + batch

or

~ Organism:treatment + batch

It depends on at which comparisons you want to look and how you view 'Organism' - i.e. confounder or covariate of interest in terms of differential expression.

Please take some time to read these Bioconductor questions and answers (one from a developer of Limma):

• Question: Removing continuous covariate effects in limma analysis

• Question: remove batch effect and adjusted model for 4 covariates in Limma

There is no correct answer here. You will be looking at this data for many weeks until you get it right.

[Kevin Blighe]

Also, apologies, here is the tutorial that I posted for 2-colour arrays: A: build the expression matrix step by step from GEO raw data

[1234anjalianjali1234]

Thankyou kevin,

I have use combat method and yes it modified my data very much, if u can provide me with some better method than combat then I will try it.

Also, I have MAlist of each sample. You have included tratment_1 only [what about different time-points]. if u don't mind, would you please to explain your codes to me? Also, I am worried about dye-swapping in few samples.

[Kevin Blighe]

Instead of using Combat, create a new column in your metadata for 'Batch', as follows (look to the right):

Sample_title                Organism Geo        protocol_1   Label_1protocol_2    label_ch2 Batch
ctrl_vs_race0_Scjm_72h_1    Fungus   GSM200000  Treated      Cy3    Control       Cy5       Batch1
ctrl_vs_race0_Scjm_72h_2    Fungus   GSM200001  Treated      Cy3    Control       Cy5       Batch1
ctrl_vs_race0_Scjm_72h_3    Fungus   GSM200002  Treated      Cy3    Control       Cy5       Batch1
ctrl_vs_race0_Scjm_96h_1    Fungus   GSM200003  Treated      Cy3    Control       Cy5       Batch1
ctrl_vs_race0_Scjm_96h_2    Fungus   GSM200004  Treated      Cy3    Control       Cy5       Batch2
ctrl_vs_race0_Stbr_72h_1    Fungus   GSM200005  Treated      Cy3    Control       Cy5       Batch2
ctrl_vs_race0_Stbr_72h_2    Fungus   GSM200006  Treated      Cy3    Control       Cy5       Batch2
ctrl_vs_race0_Stbr_72h_3    Fungus   GSM200007  Treated      Cy3    Control       Cy5       Batch2
ctrl_vs_race0_Stbr_96h_1    Fungus   GSM200008  Treated      Cy3    Control       Cy5       Batch3
ctrl_vs_race0_Stbr_96h_2    Fungus   GSM200009  Treated      Cy3    Control       Cy5       Batch3
ctrl_vs_race0_Stbr_96h_3    Fungus   GSM200010  Treated      Cy3    Control       Cy5       Batch3
ctrl_vs_complex_Scjm_72h_1  Fungus   GSM200011  Treated      Cy3    Control       Cy5       Batch3
ctrl_vs_complex_Scjm_72h_2  Fungus   GSM200012  Treated      Cy3    Control       Cy5       Batch3
ctrl_vs_complex_Scjm_72h_3  Fungus   GSM200013  Treated      Cy3    Control       Cy5       Batch3

Note that, when you refer to 'Control' for Cy5, it is not an extra control sample - it is just a 'template' DNA used on this microaray version for the purposes of calculating the log base 2 expression values of your sample in question. It is akin to the hg19 or hg38 reference genome, but it is not a true control in the sense of Case Vs. Control.

Thus, if we just focus on the Fungus samples, your design model would be:

model.matrix(formula(~protocol_1 + Batch))

Then, you could compare, for example, Treated_24h-Control_24h (please follow my tutorial step-by-step).

Just ignore dye-swapping and other organisms, for now. Keep it very simple until you feel comfortable doing basic comparisons whilst adjusting for batch. Also, again, please read the answer here: Question: Removing continuous covariate effects in limma analysis

[1234anjalianjali1234]

Thank you for your suggestions, I will try this

Actually, I have applied design model to different chips according to their experimental design, one example I have given below. By looking at my code u will know that I am comfortable doing basic comparisons. But my main problem is to combine all chips and then do a common analysis to find DEG. Also, Its very complicated as I have to consider many variables at a time [timepoints, organism, batch effect, dye-swap].

I have gone through the links provided by you. They are very informative, thankyou.

[1234anjalianjali1234]

I have writte a code for single chip, details are below:

Sample                   Organism     Geo    protocol_1  Label_1  protocol_2 label_2
ctrl_vs_1h_infested1           Insect GSM200024  T 1hr     Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested2           Insect GSM200025  T 1hr     Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested3           Insect GSM200026  T 1 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_1h_infested1swap       Insect GSM200027  T 1hr     Cy5   Control 4 hr  Cy3
ctrl_vs_1h_infested2swap       Insect GSM200028  T 1hr     Cy5   Control 4 hr  Cy3
ctrl_vs_1h_infested3swap       Insect GSM200029  T 1 hr    Cy5   Control 4 hr  Cy3
wounded_vs_wounded_spit1       Insect GSM200030  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit2       Insect GSM200031  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit3       Insect GSM200032  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wounded_vs_wounded_spit1swap   Insect GSM200033  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wounded_vs_wounded_spit2swap   Insect GSM200034  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wounded_vs_wounded_spit3swap   Insect GSM200035  WT 4 hr   Cy5   Wounded 4 hr  Cy3
ctrl_vs_4h_infested1           Insect GSM200036  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested2           Insect GSM200037  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested3           Insect GSM200038  T 4 hr    Cy3   Control 4 hr  Cy5
ctrl_vs_4h_infested1swap       Insect GSM200039  T 4 hr    Cy5   Control 4 hr  Cy3
ctrl_vs_4h_infested2swap       Insect GSM200040  T 4 hr    Cy5   Control 4 hr  Cy3
ctrl_vs_4h_infested3swap       Insect GSM200041  T 4 hr    Cy5   Control 4 hr  Cy3
wd_systemic_vs_spit_systemic1  Insect GSM200042  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic2  Insect GSM200043  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic3  Insect GSM200044  WT 4 hr   Cy3   Wounded 4 hr  Cy5
wd_systemic_vs_spit_systemic1s Insect GSM200045  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wd_systemic_vs_spit_systemic2s Insect GSM200046  WT 4 hr   Cy5   Wounded 4 hr  Cy3
wd_systemic_vs_spit_systemic3s Insect GSM200047  WT 4 hr   Cy5   Wounded 4 hr  Cy3

My Target file:

FileName    Cy3 Cy5
GSM205161.gpr   NI  NN
GSM205162.gpr   NI  NN
GSM205163.gpr   NI  NN
GSM205164.gpr   NN  NI
GSM205165.gpr   NN  NI
GSM205166.gpr   NN  NI
GSM205167.gpr   WI  WN
GSM205168.gpr   WI  WN
GSM205169.gpr   WI  WN
GSM205170.gpr   WN  WI
GSM205171.gpr   WN  WI
GSM205172.gpr   WN  WI
GSM205173.gpr   NI  NN
GSM205174.gpr   NI  NN
GSM205175.gpr   NI  NN
GSM205176.gpr   NN  NI
GSM205177.gpr   NN  NI
GSM205178.gpr   NN  NI
GSM205179.gpr   WI  WN
GSM205180.gpr   WI  WN
GSM205181.gpr   WI  WN
GSM205182.gpr   WN  WI
GSM205183.gpr   WN  WI
GSM205184.gpr   WN  WI

My code:

targets <- readTargets("target.txt")
targets1<- targetsA2C(targets)
u <- unique(targets1$Target)
f <- factor(targets1$Target, levels=u)
design <- model.matrix(~0+f)
colnames(design) <- u
corfit <- intraspotCorrelation(MA, design)
fit <- lmscFit(MA, design, correlation=corfit$consensus)
cont.matrix <- makeContrasts("(WI-WN)-(WI-NI)-(WI-NN)-(WN-NI)-(WN-NN)-(NI-NN)", levels=design)
fit2 <- contrasts.fit(fit, cont.matrix)
fit2 <- eBayes(fit2)
topTable(fit2, adjust="BH")

I am obtaining very good results with this code, up to 10 Fold Change DEG, and most of them are known to be expressed in stress.

But to complete one of my project objectives, 24 sample is not enough, I want to increase my sample size so that i can justify my objective.

[Kevin Blighe]

Yes, you are clearly adept (good) at coding. However, i'm sorry, I don't see how you can realistically merge these experiments together and extract any more useful value from it. I think that you should leave the results as independent investigations and then meta-analyse all results 'manually'. There are too many confounding factors that require adjustment and the results, from my perspective, would likely end up meaningless.

Best of luck

Kevin

[Ram]

Hello 1234anjalianjali1234!

It appears that your post has been cross-posted to another site: https://support.bioconductor.org/p/107938/

This is typically not recommended as it runs the risk of annoying people in both communities.

[1234anjalianjali1234]

Sorry for the inconvenience, how do I delete the post from Bioconductor?

[h.mon]

Now that you already posted there, I suggest you do not delete, but add a comment pointing to this thread, and apologize for the small faux pas of cross-posting.

Answer 1

[h.mon]

tl;dr: due to the confounding factors we can see, and the hidden factors we don't know, it is not advisable to combine all these experiments and analyse them together.

Long answer:

It seems to me you are trying to analyse together four different studies. It would already be difficult to analyse such a complicated design if you performed the experiment yourself and knew all the details about it. But this is public available data and, even for experiments with very detailed metadata (like those seem to be), it is very difficult to grasp all the details to properly design the statistical model. I am certain the experiments were performed at different times, which is already a complication. Do you know if all plants were the same age? Reared at the same temperature and light/dark cycle (for example, I've seen two different temperatures reading the summaries)? All plants are from the same strain? These would be taken care of (meaning, controlled for) for a single experiment, and thus would not affect it. But when you try to combine experiments, then they turn into a big issue.

A second problem is, as Kevin already pointed, the treatments and batch effects are confounded in such a way that is impossible to untangle them from each other:

1) the time points are different between the different experiments, and some experiments involve time points while others apparently don't

2) the "control" treatment is not the same for all experiments

I could point to other confounding factors, but you get the idea.

Sorry but I don't have good suggestions. It seems you are interested in investigating responses to stress, you could try to identify "stress" genes in Solanum, create gene sets and perform enrichment analyses for each experiment in separate.

[Gordon Smyth]

+1 See also my reply to the same question on Bioconductor: https://support.bioconductor.org/p/107938/#107969

[1234anjalianjali1234]

Thank you for your comment,

I am thinking to do meta-analysis instead of combined analysis.