The simulated heritabilities for the generated test data are variable, should be all the same and higher

[r03ert0]

in src/test/configtest.py the heritabilities of the simulated phenotypes are all different, and with rather low values (0.4). This makes that some analyses do not go through. Maybe we could set them all to 0.6 or 0.8, so that the analyses converge for the test despite the small N?

[ntraut]

I set all heritabilities to 0.8.
The script 00.run_all.py still crash in line 93 executing plotHSQ from 06-02.plot_hsq.R:

[1] "putamen"
[1] "thalamus"
Error in corrplot(mcor, type = "upper", method = "circle", diag = F, tl.pos = "td", :
The matrix is not in [-1, 1]!
De plus : Warning message:
Removed 242 rows containing missing values (geom_bar).
> traceback()
4: stop("The matrix is not in [-1, 1]!")
3: corrplot(mcor, type = "upper", method = "circle", diag = F, tl.pos = "td",
tl.cex = 0.8, addCoef.col = "black", number.cex = 0.7, lowCI.mat = mcor -
mse, uppCI.mat = mcor + mse, mar = c(5, 5, 5, 5)) at 06-02.plot_hsq.R#371
2: plot_hsq_bivariate(outfile_bivariate, outpdf = gsub(".txt", ".pdf",
outfile_bivariate)) at 06-02.plot_hsq.R#625

[r03ert0]

if we accept that this is not a bug, but something that may happen, we need to add a test in the code to handle the case...

[ntraut]

Maybe it is bug, at least we have to clarify why it is happening. Maybe @bitona has a clue?

[bitona]

It is not a bug, GCTA bivariates analysis converges seldomly when few samples, i found other papers reporting they could not get estimates either from this analysis. I think the plot function crashes when none of the bivariate analyses converges for a given phenotype, i will add a test to avoid the problem.

[r03ert0]

ok!