Thank you, so if I randomly select different set of genes from main list, how I can show if the scores of some list significantly higher or lower than a criteria? I thought to plot a distribution might be on the distribution I could see if sample list favours to which part of the distribution
Calculating the hypergeometric value for overlap of two lists of genes
Hi,
I have a list of genes with their biological score like below
gene score
G0295603 0.9922432
G0295719 0.9917077
G0282307 0.9876919
G0269672 0.9862853
G0284895 0.9853162
G0274031 0.9803622
>
In this list, number of genes with score> 0.5 = 2475 and number of genes with score < 0.5 =2391
If I extract 422 genes from this list, 226 genes have score > 0.5
I want to know if these 226 genes have score > 0.5 by chance or no, so
> phyper(226-1,2475,2391,422,lower.tail = FALSE)
[1] 0.1342791
Does this means, yes these 226 genes only by chance have score > 0.5 because 0.1342791 is greater than 0.05????
Thank you
• 186 views
•
link
updated
by
Jean-Karim Heriche
•
written
by
Za
•
written
by
Za
1 answer
What you're computing is the p-value of getting 226 or more genes with score >0.5 when taking 422 genes at random from your gene list. The p-value is the probability that you would get the same or greater number of genes with score >0.5 by chance when the null hypothesis is true. Typically the null hypothesis in gene enrichment is taken as the absence of association between the criterion used to extract the sample (here your 422 genes) and the characteristics of interest (here the score). So in your case, assuming that all genes regardless of score have the same chance of making it to your list of 422 genes, the probability of having 226 or more genes with score >0.5 is ~0.13. Whether this is worth drawing any conclusion is up to you. Some people would not conclude anything because this value is above some magical threshold like 0.05.
• 1 views
•
link
written
by
Jean-Karim Heriche
I am not sure I understand the question. The hypergeometric test is for categorical data represented as a contingency table e.g. number of black balls drawn from an urn that contains both white balls and black balls. In this context, you can consider that genes with score > threshold are the white balls and genes with score <= threshold are the black balls but now you seem to be asking a question about comparing scores between lists. A simple way of comparing the scores between two lists would be to compare the means. However, I don't understand where you're going with this. If you randomly draw samples, their statistics will not be different from those of the population. Maybe you should state what the question is that you're actually trying to answer and from which data.
• 1 views
•
link
written
by
Jean-Karim Heriche
In this situation, you can indeed use the hypergeometric test.
• 1 views
•
link
written
by
Jean-Karim Heriche
Log in to answer this question.
What do you mean by 'slop'?; to what does index refer?; do the same genes appear more than once in your list? Some background to your experiment would additionally be helpful.
Thanks a lot, the genes are unique.
We have a Bioconductor package for these tasks (includes visualizations): GeneOverlap
Thank you for your suggestion, do you think my interpretation is correct or wrong?