[BioC] package LIMMA design matrix for my experiment
Ryan C. Thompson
rct at thompsonclan.org
Fri Feb 15 19:20:29 CET 2013
If you're not sure how to proceed with the full experiment, I recommend
that you start with a reduced design and figure out how to model it, and
then progressively add complexity little by little. For your data, I
would start by ignoring site, year, and season, and just figure out how
to do a simple comparison of treatment vs control while dealing with the
dye effects. Once you have a reasonable design for that, you can start
figuring out how to add in the other experimental variables.
Unfortunately I can't give you any more specific advice because I have
no experience with modelling two-color arrays with limma.
-Ryan Thompson
On 02/14/2013 07:39 PM, RLW wrote:
> Dear Limma users,
>
> Your suggestions,comments, thoughts appreciated on this posting.
>
> Here is an experimental design with 2 sites (D or Z), 2 years (08 or 10), 2
> seasons (S or F), and treated/control. Samples receiving various
> combinations of these factors are placed in either Cy5 or Cy3. There are 4
> types of controls, one for each season-year: S_08_time0, F_08_time0,
> S_10_time0, and F_10_time0. Some of control samples are technical
> duplicates and marked as *_dup. See target file below.
>
> One school of thought considers this as a factorial design while another
> regards it as a nested anova (i.e., all other factors nested under one of
> the two sites). My belief is I can do neither because of a lack of common
> reference control across ALL 8 treatment combinations. While the four
> types of controls are basically the same strain/species of untreated
> organisms, they came however from different batch of lab organisms in
> different season/year so are likely different to some extent.
>
> Here are my thoughts:
> 1. analyze each treatment group against its own control separately, namely:
> D_F_08 and Z_F_08 against F_08_time0;
> D_S_08 and Z_S_08 against S_08_time0;
> D_F_10 and Z_F_10 against F_10_time0;
> D_S_10 and Z_S_10 against S_10_time0
>
> This way, the impact of season, year, and site on treated/control can only
> be inferred indirectly.
>
> 2. technical duplicates in the controls should be excluded, while all
> treated ones are unique biological samples and should be kept. This would
> make sample size unbalanced within each of eight treatment groups. Does
> this mean I have to analyze my data in single channels in order to exclude
> the technical duplicate controls from either Cy5 or Cy3? The limma user
> guide (page 88) gave an example involving a composite design (reference and
> direct comparison) where all treated samples were compared to the pooled
> control only. The authors used log ratio in that case.
>
> 3. How should we account for Cy5 vs Cy3 correlation if we are going to
> compare a few treated samples from either Cy5 or Cy3 against control
> samples also from either channel? These samples are not necessarily from
> the same arrays.
>
> Thanks for your feedback!
>
>
> Name Cy3 Cy5 Cy3SampleName Cy5SampleName
> 330_1_2 D-F-08 F-08_Time0_dup 9_22_08_D_L7 9_22_08_T0_L6
> 464_1_2 D-F-08 F-08_Time0 9_22_08_D_L6 9_22_08_T0_L3
> 331_2_2 F-08_Time0 D-F-08 9_22_08_T0_L2 9_22_08_D_L2
> 422_2_2 F-08_Time0_dup D-F-08 9_22_08_T0_L3 9_22_08_D_L3
> 423_2_4 F-08_Time0 D-F-08 9_22_08_T0_L6 9_22_08_D_L4
> 328_2_4 D-F-10 F-10_Time0_dup 9_14_10_D_L1 9_14_10_T0_L1
> 329_1_4 D-F-10 F-10_Time0 9_14_10_D_L2 9_14_10_T0_L2
> 330_1_1 D-F-10 F-10_Time0 9_14_10_D_L3 9_14_10_T0_L3
> 422_1_4 F-10_Time0_dup D-F-10 9_14_10_T0_L2 9_14_10_D_L5
> 423_1_3 F-10_Time0_dup D-F-10 9_14_10_T0_L3 9_14_10_D_L6
> 464_2_1 F-10_Time0 D-F-10 9_14_10_T0_L1 9_14_10_D_L4
> 328_1_1 D-S-08 S-08_Time0_dup 6_3_08_D_L2 6_3_08_T0_L1
> 464_1_3 D-S-08 S-08_Time0 6_3_08_D_L4 6_3_08_T0_L6
> 464_1_4 D-S-08 S-08_Time0 6_3_08_D_L5 6_3_08_T0_L8
> 331_1_4 S-08_Time0 D-S-08 6_3_08_T0_L1 6_3_08_D_L6
> 423_2_1 S-08_Time0_dup D-S-08 6_3_08_T0_L8 6_3_08_D_L8
> 329_2_2 D-S-10 S-10_Time0_dup 5_25_10_D_L2 5_25_10_T0_L3
> 330_1_3 D-S-10 S-10_Time0 5_25_10_D_L3 5_25_10_T0_L4
> 464_1_1 D-S-10 S-10_Time0 5_25_10_D_L1 5_25_10_T0_L1
> 331_2_4 S-10_Time0_dup D-S-10 5_25_10_T0_L1 5_25_10_D_L4
> 422_2_3 S-10_Time0 D-S-10 5_25_10_T0_L3 5_25_10_D_L5
> 423_2_3 S-10_Time0_dup D-S-10 5_25_10_T0_L4 5_25_10_D_L6
> 331_1_3 F-08_Time0_dup Z-F-08 9_15_08_T0_L1 9_22_08_Z_L4
> 422_2_1 F-08_Time0 Z-F-08 9_15_08_T0_L2 9_22_08_Z_L5
> 423_1_2 F-08_Time0 Z-F-08 9_15_08_T0_L3 9_22_08_Z_L7
> 328_1_3 Z-F-08 F-08_Time0 9_22_08_Z_L1 9_15_08_T0_L1
> 329_1_2 Z-F-08 F-08_Time0_dup 9_22_08_Z_L2 9_15_08_T0_L2
> 330_2_1 Z-F-08 F-08_Time0_dup 9_22_08_Z_L3 9_15_08_T0_L3
> 331_1_2 F-10_Time0_dup Z-F-10 9_7_10_T0_L4 9_14_10_Z_L4
> 422_2_4 F-10_Time0 Z-F-10 9_7_10_T0_L5 9_14_10_Z_L5
> 423_2_2 F-10_Time0_dup Z-F-10 9_7_10_T0_L6 9_14_10_Z_L6
> 328_2_1 Z-F-10 F-10_Time0 9_14_10_Z_L1 9_7_10_T0_L4
> 329_2_3 Z-F-10 F-10_Time0_dup 9_14_10_Z_L2 9_7_10_T0_L5
> 330_2_4 Z-F-10 F-10_Time0 9_14_10_Z_L3 9_7_10_T0_L6
> 331_2_1 S-08_Time0_dup Z-S-08 5_27_08_T0_L6 6_3_08_Z_L6
> 422_1_1 S-08_Time0 Z-S-08 5_27_08_T0_L7 6_3_08_Z_L7
> 328_2_3 Z-S-08 S-08_Time0 6_3_08_Z_L1 5_27_08_T0_L6
> 329_1_1 Z-S-08 S-08_Time0_dup 6_3_08_Z_L2 5_27_08_T0_L7
> 330_1_4 Z-S-08 S-08_Time0 6_3_08_Z_L3 5_27_08_T0_L8
> 423_1_1 S-10_Time0_dup Z-S-10 5_18_10_T0_L6 5_25_10_Z_L7
> 464_2_2 S-10_Time0 Z-S-10 5_18_10_T0_L3 5_25_10_Z_L5
> 464_2_4 S-10_Time0 Z-S-10 5_18_10_T0_L4 5_25_10_Z_L6
> 328_1_2 Z-S-10 S-10_Time0_dup 5_25_10_Z_L2 5_18_10_T0_L3
> 329_2_4 Z-S-10 S-10_Time0_dup 5_25_10_Z_L3 5_18_10_T0_L4
> 330_2_3 Z-S-10 S-10_Time0 5_25_10_Z_L4 5_18_10_T0_L6
>
More information about the Bioconductor
mailing list