Plot error bars in sample plot

[fedarko]

From @mortonjt's comment in #218—

Would this add in the possibility of pseudocounts? There are some scenarios where this can be very helpful. (#208)

An even better option would be to add in uncertainty via Dirichlet bootstrapping. One possibility is you perform bootstrapping (with like 100 replicates), and then pass in the means / stds. That way you can also plot error vars in the balance scatter plots to help measure the uncertainty.

Bootstrapping will look something like

import numpy as np
boots = []
table = np.array( ... ) // your biom table
pseudo = 1
for _ in range(100):
    boots.append(np.random.dirichilet(table))

boots = np.dstack(boots)
m, s = boots.mean(axis=2), boots.std(axis=2, ddof=1)

[fedarko]

Was thinking about this today, but Vega-Lite supports error bars -- see https://vega.github.io/vega-lite/docs/errorbar.html. I don't think it'd be too much work (...in the future) to change the rank or sample plot to show confidence intervals / error bars / etc., provided this data exists. The main challenge will just be testing stuff, I think.

If we have a measure of uncertainty in the differentials (as in the figure here), adjusting the rank plot to show error bars might actually be easier since we don't have to worry about formatting as much as in the sample plot.

just a thought :)

[fedarko]

Relevant to BIRDMAn; looping in @gibsramen.