The weight on enrichment score calculation

[rivers-yao]

Hi, I am trying to understand GSEA algorithm by your code .
I found that in the code functions.R line 41
es <- (abs(values) ^ power) * weight
is not like the paper's formula where a weight variable is added on .
The raw formula of P_hit(S, i) in the paper is like
es <- (abs(values ^ power) .
Am I miss something ?

[JEFworks]

Hello,

Thanks for your concern. The parenthesis make no difference in this case due to the absolute value and power being a non-negative integer. I'll update the documentation on the power variable to make this more clear.

values = seq(-10,10)
power = 3
es1 <- abs(values) ^ power
es2 <- abs(values ^ power)
es1 == es2

Feel free to let me know if you have any other questions.

Best,
Jean

[rivers-yao]

Let me make my question clear.
The weight argument does not exist in the formula [1] which is used to calculate the P_hit(S,i) in the paper while in your gsea() function a weight vector can be provided by user for each gene to calculate the P_hit(S,i) . Is that right?

Best,
Yao

[JEFworks]

Hi Yao,

Ah, thanks for the clarification. You're correct the weighted statistic is not described in the original manuscript. It is described in the GSEA user guide and is an available option in a downloaded GUI: http://www.broadinstitute.org/cancer/software/gsea/wiki/index.php/FAQ#What_is_the_difference_between_the_weighted_statistic_and_the_classic_statistic.3F_Which_should_I_use.3F

In the assessment of the significance of ES, the original manuscript also describes a full permutation of labels and recompute of all differential expression statistics ie. the ranked values, while Liger only permutes labels on the values. A good explanation can be found here: http://baderlab.org/CancerStemCellProject/VeroniqueVoisin/AdditionalResources/GSEA

Best,
Jean