Probability values (p) are the result of applying a statistical test. The essential requirement is that the experiment is replicated (at least 3 repeated measurements is recommended). So you need at least 3 expression values for a minimum of 2 different  genes, then you apply your statistical test (e.g. Student t-test, ANOVA, etc.), and from this you will obtain p values as well as other test statistics, such as F values :)

hi

https://www.biostars.org/p/16137/

https://www.informatik.hu-berlin.de/de/.../07

https://www.biostars.org/p/248858/

Probability values (p) are the result of applying a statistical test. The essential requirement is that the experiment is replicated (at least 3 repeated measurements is recommended). So you need at least 3 expression values for a minimum of 2 different  genes, then you apply your statistical test (e.g. Student t-test, ANOVA, etc.), and from this you will obtain p values as well as other test statistics, such as F values :)

Thanks Ali :)

I also agree with Dr. Andrew. Only a single point I need to add.......

When applying a statistical test in a large number of genes (considering their expression) you may get P-values with a certain level of error. To reduce the error we generally perform FDR (false discovery rate) and consider the refined P-value (called adjusted P-value).

Good idea Arindam. You can also do a statistical power test to determine the optimal sample size, which also helps to avoid type 1 errors. And you can apply corrections to post-hoc tests to also avoid errors, such as Bonferroni, which is a type of FDR test :) 

Thanks all for your useful contributions.

Thanks Andrew for your suggestion.

I performed Benjamini–Hochberg procedure for FDR in our Affymetrix gene-chip data.

The Bonferroni correction can be quite conservative as the number of comparisons increases, I think the same happens with Benjamini-Hochberg also.