Tech Notes

Q. Why do I get no genes passing the Statistical Analysis? I have 10,000 genes on my chip.

A. There are several possible causes:

1. The lack of replicate measurements or samples. When no replicates are available, GeneSpring will estimate the standard deviation using the Cross-Gene Error Model, based on deviation from 1. This estimate is very conservative and the resulting standard deviations might be large. If possible, use more replicate samples per condition, or at least replicates in some conditions.
   
   
2. Only two replicate samples per condition are available: in that case, the p-value is usually very large due to insufficient information about the variance within a replicate group.
   
   Example: The p-value is calculated based on a table with the t values (obtained from a Student's t-test, for instance) using the degrees of freedom (df, or number of replicates minus 1). With only two replicates, df=1.
   
   The table below shows t values with respect to significance levels (p-values): note that with df=1, the t value resulting from the test must be very large to obtain a highly significant p-value (i.e., must be at least 63.6 for a p-value of 0.01). With 3 replicates, (df=2), the t value needs to be much smaller to obtain the same p-value (only 9.9).
   
   
   
   To increase the statistical power for your data, increase the number of replicates per condition, whenever possible.
   
   
3. The use of a stringent multiple testing correction method in conjunction with a large gene list (for instance, the "All Genes" list). The larger the gene list, the more stringent the multiple correction will be on each gene's p-values. For more details see our analysis guide on Multiple Testing Corrections. To correct this, use a smaller gene list to start your Statistical Analysis with, and use the least stringent Multiple Testing Correction available.

1. Increase the p-value cutoff.
2. Use the Parametric test, assume variances equal, when replicate measurements are available. This test is less conservative, as it pools all variances together and thus might give results.