Implement spatial analysis

[priviere]

cf SpATS package

[priviere]

cf

• https://cran.r-project.org/web/packages/SpATS/index.html
• https://arxiv.org/abs/1607.08255

[priviere]

Done in v0.21

• add new functions that implement spatial analysis :
  • spatial.R
  • check_model.fit_model_spatial.R
  • plot.check_model_spatial.R
  • mean_comparisons.check_model_spatial.R
  • plot.mean_comparisons_model_spatial.R
• udpate the vignette

Still to do :

• test how behave the model with real data

  • how to choose between SAP() or PSANOVA() in the spatial argument ?
  • how to choose value for nseg ?
  • for other argument, I guess the default option is good !
  • is mean comparisons method ok ?

• cf cf Alternative Experimental Designs for Plant Breeding. J Zystro, M Colley, J Dawson - Plant Breeding Reviews, 2018 - books.google.com https://scholar.google.fr/scholar?cluster=4242323211746910579&hl=fr&oi=scholaralrt&hist=JsVdCA4AAAAJ:14026937503540593349:AAGBfm0BRHq-cmaHqz10vFsJdQVrKfa4Cw

• graph with ggplot instead of plot.SpATS ?

• supress message from SpATS ?

• create a data_spatial with y1, y2, y3 + update workflow with lapply in the vignette

[priviere]

How to choose between SAP() or PSANOVA() in the spatial argument ?

It seems not to change lots of things

d = data

d$col = as.numeric(as.character(d$X))
d$row = as.numeric(as.character(d$Y))
d$col_f = as.factor(d$col)
d$row_f = as.factor(d$row)

Several options:

options	spatial	random
1	"~ PSANOVA(col, row)"	"~ col_f + row_f"
2	"~ SAP(col, row)")	NULL
3	"~ PSANOVA(col, row)"	"~ col_f + row_f"
4	"~ SAP(col, row)")	NULL

Option 1 is the one used in the example in the article https://arxiv.org/abs/1607.08255 page 33 of the pdf

m1 = SpATS(
  response = "intervalle_floraison_dj", 
  genotype = "germplasm", 
  genotype.as.random = TRUE,
  spatial = ~ PSANOVA(col, row),
  random = ~ col_f + row_f, 
  data = d)
s1 = summary.SpATS(m1, which = "variances")$p.table.vc

m2 = SpATS(
  response = "intervalle_floraison_dj", 
  genotype = "germplasm", 
  genotype.as.random = TRUE,
  spatial = ~ PSANOVA(col, row),
  data = d)
s2 = summary.SpATS(m2, which = "variances")$p.table.vc

m3 = SpATS(
  response = "intervalle_floraison_dj", 
  genotype = "germplasm", 
  genotype.as.random = TRUE,
  spatial = ~ SAP(col, row),
  random = ~ col_f + row_f, 
  data = d)
s3 = summary.SpATS(m3, which = "variances")$p.table.vc

m4 = SpATS(
  response = "intervalle_floraison_dj", 
  genotype = "germplasm", 
  genotype.as.random = TRUE,
  spatial = ~ SAP(col, row),
  data = d)
s4 = summary.SpATS(m4, which = "variances")$p.table.vc

Look at link between BLUPs estimation

plot(m1$fitted, m2$fitted)
plot(m1$fitted, m3$fitted)
plot(m1$fitted, m4$fitted)
plot(m2$fitted, m3$fitted)
plot(m2$fitted, m4$fitted)
plot(m3$fitted, m4$fitted)

Regarding variznces, variances are more important on germplasm when row and col are not in random

c(s1["germplasm", "Variance"], s2["germplasm", "Variance"], s3["germplasm", "Variance"], s4["germplasm", "Variance"])

For residual variance, it changed nothing

c(s1["Residual", "Variance"], s2["Residual", "Variance"], s3["Residual", "Variance"], s4["Residual", "Variance"])

So lets choose model m1

[priviere]

how to choose value for nseg ?
for other argument, I guess the default option is good !

Lets use the default arg ...

[priviere]

Done in v0.25

• update mean comparisons
• supress message from SpATS ? Not done, not useful I guess
• create a data_model_spatial with y1
• update documentation

[priviere]

Regarding mean comparisons, cf #124