Very simple statistic question

[potatoes]

If I want to know whether signal of protein X at protein Y peaks is significantly/not significant higher/lower value than random regions in the genome,

I have 1000 peaks of protein Y with length 1000bp in mouse genome and calculate the average signal of protein X ChIP-seq at these peaks
As control I shuffle these peaks 1000 times randomly in the genome and calculate the average signal of protein X ChIP-seq at these shuffled peaks

Therefore I have observation table:
Peak Signal
Peak_A 100
Peak_B 150
Peak_C 50
etc

And shuffle table:
Peak Signal_1 Signal_2 Signal_3 ... Signal_1000
Peak_A 99 94 14 ... 50
Peak_B 50 90 100 ... 25
Peak_C 25 60 35 ... 40

What is the best method and how to calculate p-value? Why?

Thank you

[potatoes]

200 views and no one answer?

I thought this would be an easy one... I was going to use Monte Carlo to find p-value but someone else suggested Wilcoxon, I'm just not sure which is correct

[dpryan]

Well, the shuffling might suffice in place of a Monte Carlo. The p-value for each peak ends up being equal to the number of shuffled peaks with greater or equal signal divided by the number of shuffled iterations. This is also how you'd get a p-value if you were to do a Monte Carlo simulation. Obviously, your p-value is limited by the number of iterations performed, so the more you do the more precise your results.

[potatoes]

Well, the shuffling might suffice in place of a Monte Carlo. The p-value for each peak ends up being equal to the number of shuffled peaks with greater or equal signal divided by the number of shuffled iterations. This is also how you'd get a p-value if you were to do a Monte Carlo simulation. Obviously, your p-value is limited by the number of iterations performed, so the more you do the more precise your results.

Thank you. I agree that is how I should calculate p-value. But since I want to know at a whole whether protein X correlate with protein Y in the genome, and not which peak is significantly correlated, how should I calculate the final p value though?

If I get p-value for each peak, I'm not sure how to convert them to a final p-value. I can't just average p-value right? Or can I make that claim if the majority of peak (e.g. 80% ) has significantly higher signal than shuffled (p < 0.05)?

[dpryan]

Ah, I thought you wanted per-peak p-values. I would just do the simulation then and compare the distributions of distances between peaks of the two proteins (or just bin everything and then compare bin occupancy).

[potatoes]

Alright, Thank you!