Statistically validate number of exons and isoforms

[int11ap1]

Hello to everybody,

I was reading the following paper,

Derrien, T; Johnson, R; Bussotti, G; Tanzer, A; Djebali, S; Tilgner, H; Guernec, G; Martin, D; Merkel, A; Knowles, DG; Lagarde, J; Veeravalli, L; Ruan, X; Ruan, Y; Lassmann, T; Carninci, P; Brown, JB; Lipovich, L; Gonzalez, JM; Thomas, M; Davis, CA; Shiekhattar, R; Gingeras, TR; Hubbard, TJ; Notredame, C; Harrow, J; Guigó, R (Sep 2012). "The GENCODE v7 catalog of human long noncoding RNAs: analysis of their gene structure, evolution, and expression.". Genome Research 22 (9): 1775–89

which has the following figure:



For B and C figures, they apply a t-test in order to know if coding and lncRNAs have significativelly different exon, intron and transcript lengths. However, in figures A and D, they do not mention any statistical analysis. Might that data (figures A and D) be analyzed using another t-test?

Thanks.

[dpryan]

No, and they shouldn't have used a T-test with the other results in that figure (they should have used a Wilcoxon test).

Subfigures A and D should be analyzed with ordered logistic regression.

[int11ap1]

No, and they shouldn't have used a T-test with the other results in that figure (they should have used a Wilcoxon test).

Because they are not normally distributed, right? What if I apply a log transformation? They should be nearly normal then.

Could you be more specific about ordered logistic regression, please?

Thanks for answering.

[dpryan]

Yeah, log transforming them might make them normal enough.

I'm not sure how much more specific I could be about ordered logistic regression. You can do that easily in R, if that's what you want to know.

[int11ap1]

Yeah, log transforming them might make them normal enough.

I'm not sure how much more specific I could be about ordered logistic regression. You can do that easily in R, if that's what you want to know.

I do not understand how could I compare two populations just doing a regression...

Another question, if I perform a Wilcoxon test for the other data and the two populations are significantly not the same, should I look at the median to see what population is more expressed?

[dpryan]

A T-test is a regression. An Anova is a regression. Most of the statistics you've ever done have been regressions, even if you didn't know it (in fact, a T-test and an ANOVA are just special cases of linear models).

Regarding the Wilcoxon, yes, looking at the median would be a good way to go.