This web page was produced as an assignment for a course on
Statistical Analysis on Microarray Data at
Description of
specific experimental design of the study
These experiments were conducted on microarrays. They call
Channel 1 = Green
Channel 2 = Red
This is a reference design in which all of the samples were in Channel 2 (red) and the reference in Channel 1 (green) is something called UHR. The spots on the arrays represented genes from humans, mice, rats, predicted human miRNAs, and others labeled control.
Comment on deign,
variability, references, randomization, replication, fixed vs
random effects
Reference design was appropriate for this study because it allows for easy comparison of the different samples.
Sources of variability arise from technical variability (different microarrays are inherently put under different conditions), biological variability (since there will be variability between tissue samples of various individual), and genetic variability (which is what they want to detect). To lessen technical variability spots were printed in duplicate, and to lessen biological variability some of the conditions of interest had several samples taken.
Each microarray is made of a bunch of smaller grids of genes 8 across and 6 down. The top 3 of these rows are repeated again in the bottom three rows. For example, here is array 30 (76166) (If the links do not work correctly, go here and select “display data” and then the links below should work) where you can see how they are duplicated. Unfortunately this results in the spot duplicates being very close together so effects that are problems on an isolated part of the array will end up affecting both spots instead of just one. As described in the “interesting information about my project” section, there were several arrays where there were problems in background noise that is more in some areas than others. This method of spot duplication was not as effective as it could have been.
There is also biological replication in this experiment: there are 8 GIST, 7 SS, 6 LMS, 1 DDLPS, 3 ARMS, 2 PRMS, 1 ERMS, 5 NSM, and 2 normal skeletal muscle samples. Except for the DDLPS and the ERMS, there replication on a biological level. The ARMS, PRMS, and ERMS are all subgroups of RMS.
Fixed effects include the effects of the genes between different types of tissue sample, and the effect of the dyes on the RNA (the effect of the dye is fixed since it is red for all different samples and green for the reference).
Random effects include the biological variability of the samples, and the effects of being on different arrays.
How the design might
be improved upon if they were to do it again
Because a fair number of the microarrays in this experiment are of poor quality (again see interesting information), the duplication of spots is important. Unfortunately, the duplicating that they do does very little to correct for this since the duplicated spots are so near each other. They should have had duplicate spots on different parts of the array.
Also, it would have been good if they had produced consistently high quality arrays.
The article seems to be written for a knowledgeable audience which already knows about the importance of the different tissue samples (GIST, SS, LMS, DDLPS, ARMS, PRMS, ERMS, NSM, normal skeletal). Perhaps there is more variability in GIST and SS than the other ones which is why there are more cases of those, but to people like me who do not know, it seems as though it would have made more sense to have the same number of samples for each of these conditions.
This website was designed by Austen Head.
Email: austen [dot] head [at]