Comparing gaussian distributions with Gumbel-like distribution

AI Thread Summary
The discussion centers on analyzing the differences between a Gaussian distribution of control signals and a Gumbel-like distribution observed during analyte binding events across 10,000 sensors. The initial control experiment yields a narrow Gaussian distribution due to electronic noise, while the analyte measurements show a distinct Gumbel distribution. Participants suggest conducting normality tests, such as the Shapiro-Wilk test, to verify the Gaussian nature of the control data. For comparing the two distributions, the two-sample Kolmogorov-Smirnov test is recommended, along with Wilcoxon's rank-sum test for assessing mean differences. This approach will provide a quantitative analysis of the distribution differences.
TryingTo
Messages
6
Reaction score
0
Hi all,

I study binding of analytes in a platform where I have 10.000 sensors. Theres is one binding event per sensor and I identify it as a sudden positive change in the signal. I do first a control experiment without analytes. I measure the maximum change in the signal for each sensor and I obtain a narrow gaussian distribution around cero due to electric noise (green curve). When I measure the analytes I obtain a kind of Gumbel distribution because some sensors detect a positive binding event (larger than the electronic noise, red curve). When I compare the histograms is clear that there is a difference before and after but I would like to do a quantitative analysis of how different the distributions are. Do you have any clue on how to do this? Which test I could apply? One of the distributions is normal but the other is not so I'm not sure.

Thank you!
Untitled-1.png
 
Physics news on Phys.org
TryingTo said:
Hi all,

I study binding of analytes in a platform where I have 10.000 sensors. Theres is one binding event per sensor and I identify it as a sudden positive change in the signal. I do first a control experiment without analytes. I measure the maximum change in the signal for each sensor and I obtain a narrow gaussian distribution around cero due to electric noise (green curve). When I measure the analytes I obtain a kind of Gumbel distribution because some sensors detect a positive binding event (larger than the electronic noise, red curve). When I compare the histograms is clear that there is a difference before and after but I would like to do a quantitative analysis of how different the distributions are. Do you have any clue on how to do this? Which test I could apply? One of the distributions is normal but the other is not so I'm not sure.Thank you!

Hey!

First of all, I'm curious as to what led you to think that the measurements of the analytes experiment follow a kind of Gumbel distribution. Were you given prior information stating that a Gumbel distribution was to be expected or did you just assume it followed that distribution by looking at its shape? Also, how confident are you that the data of the control experiment follow a normal distribution? I would definitely start by testing that. You can use the Shapiro-Wilk test for normality or you can compare the data of the control experiment with a normal distribution using the one-sample Kolmogorov-Smirnov test. You can also use the two-sample K-S test to compare both samples and see if they're significantly different. Finally, if you want to compare the means of both experiments and see if there's a significative difference, you can use a nonparametric test like Wilcoxon's rank-sum test, since I believe normality can't be assumed for at least one of both samples.

I hope this helps!
 
Back
Top