0
$\begingroup$

I have the following dataframe that is comprised of enzyme substrates as the variable and then the unique number of times enzymes capiable of degrading this substrates appeared in my genomes and then tot total count of times all of those enzymes appeard in my samples. For example, a 7 unique enzymes were found to be capiable of degrading chitin, and they occured a total of 298 times across my genomes.

I wish to see if any particular substrate has significantly larger occurance in both unique and total, than the rest of the substrates. Would outlier detection be appropriate for this or would a simple T-test do?

enter image description here

$\endgroup$
0
0
$\begingroup$

I agree you have the right idea because that data doesn't look at all linear, so regression analysis looks really complicated. Your approach a straight comparison keeps it simple and is within the biological question you are asking,

However, I would use a non-parametric such as Wilcoxon instead of a T-test. I don't fully know the null distribution you are comparing it against. Wilcoxan doesn't require an assumption of normality. T-tests invariably require a transformation to normality and that gets complicated, but a log or sqrt transformation might be okay. If Wilcoxan produces a no significant difference you report it and move onto a T-test.

It is difficult to use outlier analysis with 9(?) observations, thats used if you have a way-out result and can afford to exclude it (>3 standard deviations). Outlier analysis is to remove a single result that is messing up your analysis and doesn't affect the bulk of the data. You alot of observations to be considering outlier analysis.

$\endgroup$
5
  • $\begingroup$ The actual dataset is 130 substrates thankfully. But I still think you are correct that it would not be normally distributed. Thus would a glm work here with either a negative binomial or poisson as the distribution? $\endgroup$
    – Lamma
    Jun 10 '20 at 6:35
  • $\begingroup$ I think negative binomial would be best as I don't think the variance is equal to the mean, but I guess Icould just test that assumption. $\endgroup$
    – Lamma
    Jun 10 '20 at 6:43
  • $\begingroup$ That would need fitting and at thiss point you are into formal regression analyses $\endgroup$
    – M__
    Jun 10 '20 at 10:01
  • $\begingroup$ I have changed the analysis now to have the frequence of occurance of each enzyme and lables them via the enzyme substrate so my model is glm.nb(substrate ~ frequency). I think this is better as it generates a mean and variance for each substrate thus allowing a much better identification of if the degridation of any singular substrate is enriched? $\endgroup$
    – Lamma
    Jun 10 '20 at 10:06
  • $\begingroup$ Okay thats a different question $\endgroup$
    – M__
    Jun 10 '20 at 10:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.