I have a microarray data set of genes with the expression values in 3 replicates.
I would like to select the genes whose expression value is high and consistent across all experiments (the standard deviation should be less). I am searching for the best statistical method to solve my issue.
Please see below for an example:
ID Experiment 1 - Experiment 2 - Experiment 3
Gene 1 - 10 11 8
Gene 2 - 12 10 11
Gene 3 - 4 12 12
Gene 4 - x x x
Gene 1000 - x x x
Perhaps calculate : The coefficient of variation of each gene: the (CV) is defined as the ratio of the standard deviation to the mean. It shows the extent of variability in relation to mean of the population
A nonparametric possibility is the quartile coefficient of dispersion, i.e. interquartile range divided by the median .
Dear Asha,
You may find the genefilter Bioconductor package useful (http://www.bioconductor.org/packages/release/bioc/html/genefilter.html).
The package allows you to filter genes according to expression (e.g. high expression) and variance (e.g. low variance across conditinos ~ consistent expression)
best wishes,
Ignat
The Bioconductor package mentioned above is indeed useful. If you are looking for an easy way in Excel. You could calculate mean expression for each gene and set an arbitrary cut-off for expression to only select highly expressed genes. Subsequently, you could calculate the interquartile range (for example =PERCENTILE(A1:C1,0.75)-PERCENTILE(A1:C1,0.25) for each row as measure for variability. The ones with the lowest IQR are the most consistently expressed.
Perhaps calculate : The coefficient of variation of each gene: the (CV) is defined as the ratio of the standard deviation to the mean. It shows the extent of variability in relation to mean of the population
A nonparametric possibility is the quartile coefficient of dispersion, i.e. interquartile range divided by the median .
The above suggestions are all fine, but your question is not really clear. Do you have 3 replicates of the the same sample? If so there should be little variability in your experiment. Any observed variability will be some mix of technical/biological replication. If you have multiple sets of triplicates for a treated and untreated cell line, for example, and looked at low variability then you measuring genes with little change regarding the specific treatment. However another treatment might affect the genes. You only know the gene's variability in regard to the experimental conditions tested. Please elaborate more if I have not understood you problem.
Bioconductor and R is the best !!! GeneSpring also has both the options of filtering the highly expressed genes and to perform the statistical analysis !!!!!
Please see enclosed excel sheet showing the results for your example using Rank order and IQR methods. If you are looking for most highly expressed gene in all three replicates with possible low variability, then both tests will give same answer i.e. Gene 2. But, if you prefer to look for most consistent (i.e. less variable) with subsequent high expression, then better to use rank order with first sorting on Rank of SD followed by Rank of Mean (largest to smallest in both conditions). In this case, Gene 1 is the answer.
Than you everyone for the suggestion..!!
Now i will try each method and compare the outcome....Thanks once again...!
You have the answer here
http://onlinelibrary.wiley.com/doi/10.1002/1873-3468.12192/abstract
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Gene Expression