Question

Comparing discrete variable to PCA

0

Entering edit mode

k.vitting.seerup ▴ 120

@kvittingseerup-7956

Last seen 16 months ago

European Union

Is there a Bioconductor solution of comparing discrete variables to PCs to figure out associations. I'm looking for the discrete version of PCAtools' eigencor plot.

Cheers Kristoffer

EDA pcaExplorer PCAtools PCA • 2.0k views

ADD COMMENT • link updated 3.0 years ago by Miguel ▴ 20 • written 4.3 years ago by k.vitting.seerup ▴ 120

score 1 · Answer 1 · 2020-08-19

1

Entering edit mode

Kevin Blighe ★ 4.0k

@kevin

Last seen 6 weeks ago

Republic of Ireland

Hey Kristoffer, I developed PCAtools, as you know - is eigencorplot not what you need? PCAtools has been in Bioconductor for > 1 year.

Note that these are the exact same:

Pearson correlation coefficient of Continuous X versus Categorical Y, with Y encoded numerically
Extracting the r correlation value from a linear regression of the form X ~ Y

As I show here:

continuous <- c(45, 67, 12, 65, 75, 3, 44, 90)
categorical <- factor(c(0,0,0,0,1,1,1,1))

cor(continuous, as.numeric(categorical)) ^ 2
[1] 0.01024737

summary(lm(continuous ~ categorical))$r.squared
[1] 0.01024737

[source: https://www.biostars.org/p/349397/#349493]

Kevin

ADD COMMENT • link 4.3 years ago Kevin Blighe ★ 4.0k

0

Entering edit mode

Hi Kevin

I did try eigencorplot() but got this error when using a categorical (text or factor) variable:

Error in cor(xvals, yvals, use = corUSE, method = corFUN) : 
  'y' must be numeric
In addition: Warning message:
In eigencorplot(myPca, metavars = c("barcode")) :
  barcode is not numeric - please check the source data as everything will be converted to a matrix

Turns out it can be solved by convert them to numerical values before using pca().

Thanks for pointing out it could be done.

Cheers Kristoffer

ADD REPLY • link 4.3 years ago k.vitting.seerup ▴ 120

1

Entering edit mode

Oh, I thought that issue was addressed in the previous Bioc release, i.e., it should automatically convert factors to numeric and give a warning, like above. I have not yet come across the other error thrown by cor() internally.

ADD REPLY • link 4.3 years ago Kevin Blighe ★ 4.0k

0

Entering edit mode

Would that same approach hold true when we have multiple categories.

Let's say ''group_1", ''group_2" and ''group_3" In this case, each will be assigned to an integer (0,1,2) although they are ordered categories they do not exactly match this transformation. Am I right?

If so, any suggestions to working with multiple categories variabels?

ADD REPLY • link 3.0 years ago Miguel ▴ 20