#May 2021
library("tidyverse")
library("lme4")
library("effects")
library("arm")
library("pbkrtest")

setwd("") #your working directory here
load("Belyk_EBR_S6h_preEMG.Rdata")


######################
###custom functions###
######################
#preferred ggplot themes
theme_mb<-function(){
  theme_light(base_size=16) %+replace% 
    theme(strip.background = element_rect(fill="grey95"),
          strip.text.x = element_text(colour="black"),
          strip.text.y = element_text(colour="black", angle=270),
          axis.text.x = element_text(angle = 35, vjust = 1, hjust=1))
}
##################
####some plots####
##################
#logs contains newly calculated observations based 50ms of EMG signal before stimulation
logs %>% #filter(Participant >9) %>%
  mutate(Coil = as.factor(Coil)) %>%
  mutate(Coil = relevel(Coil, "Test")) %>%
  ggplot(aes(x=PreEMG_uVS, fill=Condition))+
  geom_density()+
  facet_grid(Condition~Coil)+
  xlab("Pre-stimulus EMG (uV/S)")+
  ggtitle("Distributions of prestimulation EMG")+
  xlim(0,100)+
  theme_mb()
ggsave("Verify_precontraction_state.pdf")

logs %>% #filter(Participant >9) %>%
  ggplot(aes(x=PreEMG_uVS, y=MEP, color=Condition))+
  geom_point()+
  geom_smooth(method="lm")+
  facet_grid(Condition~Coil)+
  xlab("Pre-stimulus EMG (uV/S)")+
  xlim(0,100)+
  theme_mb()

PreEMGmod <- glmer(PreEMG_uVS ~ Coil*Condition + (1+Coil*Condition|Participant), family=Gamma(link="log"), data=logs)
relgrad <- with(PreEMGmod@optinfo$derivs,solve(Hessian,gradient)) #is this less than about 0.0001? pushing it slightly
max(abs(relgrad)) #good enough but not ideal
summary(PreEMGmod)

####################
###Residual check###
####################
plot(residuals(PreEMGmod)) #ok, a little funk at the lower tail but comprising few observations
qqnorm(residuals(PreEMGmod)) #a few weirdos at the bottom tail

#roughly these ones
logs %>%
filter(residuals(PreEMGmod)< -3) 

######################################################################
###contrasts for comparing rest in RR vs AR and active in  AA vs AR###
######################################################################

f<-expand.grid(unique(logs$Coil), unique(logs$Condition))
names(f)<-c("Coil", "Condition")
contrast_matrix<-model.matrix(~Coil*Condition,f)

contrast_matrix<-as_tibble(contrast_matrix)
contrast_matrix<-cbind(f, contrast_matrix)
contrast_matrix<-logs
contrast_matrix$ConRest=0
contrast_matrix$ConActive=0
contrast_matrix<-contrast_matrix %>%
  mutate(ConRest= replace(ConRest,Coil=="Conditioning" & Condition == "RR", 0.5)) %>%
  mutate(ConRest= replace(ConRest,Coil=="Conditioning" & Condition == "AR", -0.5)) %>%
  mutate(ConActive= replace(ConActive,Coil=="Test" & Condition == "AA", 0.5)) %>%
  mutate(ConActive= replace(ConActive, Coil=="Test" & Condition == "AR", -0.5))

PreEMGmodcons<-glmer(PreEMG_uVS ~1+ ConRest+ConActive + (ConActive|Participant),
                     family=Gamma(link="log"), data=contrast_matrix)
summary(PreEMGmodcons)

############################
###estimates for plotting###
############################
PreEMGmod.effects<-Effect(c("Coil","Condition"),PreEMGmod, confint=T)
preds<-PreEMGmod.effects$x
preEMG.Gamma<-PreEMGmod.effects$fit
PreEMGmod.plotalble<-cbind(PreEMGmod.effects$x,PreEMGmod.effects$fit,PreEMGmod.effects$lower,PreEMGmod.effects$upper)
lower<-PreEMGmod.effects$lower
upper<-PreEMGmod.effects$upper
PreEMGmod.plotalble<-as_tibble(cbind(preds,preEMG.Gamma, lower,upper))
rm(preds, preEMG.Gamma, lower, upper)

PreEMGmod.plotalble<-PreEMGmod.plotalble %>%
  mutate(Coil = fct_relevel(Coil, "Test")) #change order of levels for plotting

ggplot(PreEMGmod.plotalble, aes(x=Coil, y=preEMG.Gamma, color=Condition))+
  geom_point(size=2)+
  geom_linerange(aes(ymin=lower,ymax=upper))+
  facet_grid(.~Condition)+
  ylab("Model coefficients")+
  ggtitle("Pre-stimulus EMG")+
  theme_mb()
ggsave("Pre_EMG_gamma.pdf")


######################
###DIFFERENCE MODEL###
######################
#attempt to run my original models again
#dat2b_preEMG adds the newly computed prestim EMG to the original model data
#data preEMG from 2 Subjects was missing. This dataframe has data for those participants removed 
#because bootsrapping with PBmodcomp cannot cope with missing data

mod_inhibit_dif_preEMG<-lmer(Inhibition ~  #including all covariates here
                               IHI+Condition*Cond_EMG #dropped IHI interaction. It caused fit problems and was not preffered by AIC anyway
                             + Cond_Target_Norm +Cond_Twist_Norm+Cond_Angular_Norm
                             + Test_Target_Norm +Test_Twist_Norm+Test_Angular_Norm
                             + PreEMG_Test + PreEMG_Conditioning
                             + (1|Subject)+(0+Cond_EMG|Subject)
                             ,REML=F
                             ,control =lmerControl(optimizer = "bobyqa", optCtrl=list(maxfun=1e5))  #fiddle with optimizers
                             ,data=dat2b_preEMG)
relgrad <- with(mod_inhibit_dif_preEMG@optinfo$derivs,solve(Hessian,gradient)) #is this less than about 0.0001? If so can probably ignore convergence warnings
max(abs(relgrad)) #good enough

mod_inhibit_dif_preEMG_stnd<-standardize(mod_inhibit_dif_preEMG)
summary(mod_inhibit_dif_preEMG_stnd)
mod_inhibit_dif_preEMG_stnd_CI<-confint(mod_inhibit_dif_preEMG_stnd)

pdf("plots/Diff_QQ.pdf") #pretty good
qqnorm(scale(residuals(mod_inhibit_dif_preEMG_stnd))) #something weird here, but should be fine with bootstrapping
abline(a=0,b=1)
dev.off()

pdf("plots/Diff_dispersion.pdf") #pretty good
plot(mod_inhibit_dif_preEMG_stnd, main="Dispersion of Differences Model") #pretty good
dev.off()

###BOOTSTRAPPING###
#reduced interactions
#I = IHI, C = Condition, E = CS MEP

mod_inhibit_dif_preEMG_mains<-update(mod_inhibit_dif_preEMG_stnd, .~. -Condition:z.Cond_EMG)
mod_inhibit_dif_preEMG_I<-update(mod_inhibit_dif_preEMG_mains, .~. -c.IHI)
mod_inhibit_dif_preEMG_C<-update(mod_inhibit_dif_preEMG_mains, .~. -Condition)
mod_inhibit_dif_preEMG_E<-update(mod_inhibit_dif_preEMG_mains, .~. -z.Cond_EMG)

set.seed(42) #for replicability
(PB_CE_dif_preEMG<-PBmodcomp(mod_inhibit_dif_preEMG_stnd, mod_inhibit_dif_preEMG_mains,  nsim=5000))
(PB_I_preEMG_dif<-PBmodcomp(mod_inhibit_dif_preEMG_mains, mod_inhibit_dif_preEMG_I,  nsim=5000))
(PB_C_preEMG_dif<-PBmodcomp(mod_inhibit_dif_preEMG_mains, mod_inhibit_dif_preEMG_C,  nsim=5000))
(PB_E_preEMG_dif<-PBmodcomp(mod_inhibit_dif_preEMG_mains, mod_inhibit_dif_preEMG_E,  nsim=5000))