#Primary hypothesis, soc & phys regions vs. func V1
roi_list_primary_R <- c("RSMG", "RSPC","RTPJ","RSTS","funcRV1", "anatRV1")
roi_list_primary_L <- c("LSMG", "LSPC","LTPJ","LSTS","funcLV1","anatLV1")
fc_primary_R <- get_fc_table(all_subjects, roi_list_primary_R, by_barrier=FALSE)
fc_primary_L <- get_fc_table(all_subjects,roi_list_primary_L,by_barrier=FALSE)fc_analysis
Single run, single subject
All subjects
soc & phys blocks
Visualization
#left hemisphere
ggplot(subset(fc_primary_L, trial_type %in% c("social", "physics")), aes(x = ROI_pair, y=correlation,fill=trial_type))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "coral2"))+
theme_minimal()+
labs(title="Left Hemisphere")+
theme(plot.title = element_text(size=15))+
ylim(-0.5,1)+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
#right hemisphere
ggplot(subset(fc_primary_R, trial_type %in% c("social", "physics")), aes(x = ROI_pair, y=correlation,fill=trial_type))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "coral2"))+
theme_minimal()+
labs(title="Right Hemisphere")+
theme(plot.title = element_text(size=15))+
ylim(-0.5,1)+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
Analysis
Take a pair of regions, one soc, one phys, see if they’re more connected during social than physical task. This connection is specific to this pair, don’t see it when you pair up either region with V1.
Simple Effect Model
compare_pair <- function(roi_list,roi_pair) {
df <- roi_list %>%
filter(ROI_pair == as.character(roi_pair)) %>%
filter(trial_type != "rest") %>%
mutate(subject = str_remove(subject_run, "_.*$"))
df$trial_type <- relevel(as.factor(df$trial_type), ref = "physics")
model <- lmer(correlation ~ trial_type + (1|subject), data = df)
return(summary(model))
}
#fc_primary_L: LSMG, LSPC, LTPJ, LSTS, funcLV1
#fc_primary_R: RSMG, RSPC, RTPJ, RSTS, funcRV1
compare_pair(fc_primary_R,"RSPC_RSTS")Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: correlation ~ trial_type + (1 | subject)
Data: df
REML criterion at convergence: -18.1
Scaled residuals:
Min 1Q Median 3Q Max
-1.86106 -0.61502 -0.09085 0.58278 2.01820
Random effects:
Groups Name Variance Std.Dev.
subject (Intercept) 0.02287 0.1512
Residual 0.02566 0.1602
Number of obs: 56, groups: subject, 14
Fixed effects:
Estimate Std. Error df t value Pr(>|t|)
(Intercept) 0.11599 0.05050 19.02889 2.297 0.0331 *
trial_typesocial 0.07966 0.04281 41.00000 1.861 0.0700 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Correlation of Fixed Effects:
(Intr)
tril_typscl -0.424RSPC-RSTS are more connected during soc than phys trials, the difference is marginally significant (b=0.080, p=.07). We don’t see this pattern in their connectivity with functional RV1.
compare_pair(fc_primary_R,"RSPC_anatRV1")
compare_pair(fc_primary_R,"RSMG_anatRV1")
compare_pair(fc_primary_L,"LSPC_anatLV1")But LSPC and anat LV1 have significantly higher connectivity during soc than phys trials (b=0.11, p<.01), same for RSPC and anat RV1 (b=0.082, p=.01), and RSMG and anat RV1 (b=0.078, p<.01). These are unexpected.
None of the other pairs had significant difference in soc vs. phys trials.
Interaction Model
the effect of trial_type is greater for soc-phs pair, than soc-V1 pair and phys-V1 pair
(if you want to interpret main effects properly in this model, summed contrasts) - but dummy coding can better answer the question about difference between roi pairs?
library(emmeans)Welcome to emmeans.
Caution: You lose important information if you filter this package's results.
See '? untidy'interaction <- function(data, roi_set) {
df <- data %>%
filter(ROI_pair %in% roi_set) %>%
filter(trial_type != "rest") %>%
mutate(subject = str_remove(subject_run, "_.*$"),
ROI_pair = factor(ROI_pair, levels = c(roi_set[1], setdiff(roi_set, roi_set[1]))))
model <- lmer(correlation ~ ROI_pair * trial_type + (1 | subject), data = df)
print(summary(model))
emm <- emmeans(model, ~ trial_type | ROI_pair) # estimated marginal means of correlation for each level of trial_type, within each ROI_pair, accounting for interaction (same coefficient but dif p value than simple effect model)
print(contrast(emm, method="pairwise")) #Are predicted correlations for soc vs. phys trials different within the same ROI_pair?
}
interaction(fc_primary_R, roi_set = c("RSPC_RSTS", "RSPC_funcRV1", "RSTS_funcRV1"))Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: correlation ~ ROI_pair * trial_type + (1 | subject)
Data: df
REML criterion at convergence: -122.8
Scaled residuals:
Min 1Q Median 3Q Max
-2.3453 -0.5993 0.1031 0.6683 2.5751
Random effects:
Groups Name Variance Std.Dev.
subject (Intercept) 0.01122 0.1059
Residual 0.02062 0.1436
Number of obs: 168, groups: subject, 14
Fixed effects:
Estimate Std. Error df t value
(Intercept) 0.11599 0.03922 34.66996 2.958
ROI_pairRSPC_funcRV1 0.33110 0.03838 149.00000 8.627
ROI_pairRSTS_funcRV1 0.12054 0.03838 149.00000 3.141
trial_typesocial 0.07966 0.03838 149.00000 2.075
ROI_pairRSPC_funcRV1:trial_typesocial -0.09290 0.05428 149.00000 -1.711
ROI_pairRSTS_funcRV1:trial_typesocial -0.08338 0.05428 149.00000 -1.536
Pr(>|t|)
(Intercept) 0.00555 **
ROI_pairRSPC_funcRV1 8.73e-15 ***
ROI_pairRSTS_funcRV1 0.00203 **
trial_typesocial 0.03967 *
ROI_pairRSPC_funcRV1:trial_typesocial 0.08908 .
ROI_pairRSTS_funcRV1:trial_typesocial 0.12664
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Correlation of Fixed Effects:
(Intr) ROI_pRSPC_RV1 ROI_pRSTS_RV1 trl_ty ROI_RSPC_RV1:
ROI_pRSPC_RV1 -0.489
ROI_pRSTS_RV1 -0.489 0.500
tril_typscl -0.489 0.500 0.500
ROI_RSPC_RV1: 0.346 -0.707 -0.354 -0.707
ROI_RSTS_RV1: 0.346 -0.354 -0.707 -0.707 0.500
ROI_pair = RSPC_RSTS:
contrast estimate SE df t.ratio p.value
physics - social -0.07966 0.0384 149 -2.075 0.0397
ROI_pair = RSPC_funcRV1:
contrast estimate SE df t.ratio p.value
physics - social 0.01324 0.0384 149 0.345 0.7306
ROI_pair = RSTS_funcRV1:
contrast estimate SE df t.ratio p.value
physics - social 0.00372 0.0384 149 0.097 0.9229
Degrees-of-freedom method: kenward-roger interaction(fc_primary_R, roi_set = c("RSPC_RSTS", "RSPC_anatRV1", "RSTS_anatRV1"))Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: correlation ~ ROI_pair * trial_type + (1 | subject)
Data: df
REML criterion at convergence: -100.2
Scaled residuals:
Min 1Q Median 3Q Max
-2.4642 -0.5870 -0.0035 0.6622 2.2736
Random effects:
Groups Name Variance Std.Dev.
subject (Intercept) 0.01692 0.1301
Residual 0.02322 0.1524
Number of obs: 168, groups: subject, 14
Fixed effects:
Estimate Std. Error df t value
(Intercept) 0.115987 0.045142 29.099441 2.569
ROI_pairRSPC_anatRV1 0.224757 0.040728 149.000000 5.519
ROI_pairRSTS_anatRV1 0.249402 0.040728 149.000000 6.124
trial_typesocial 0.079658 0.040728 149.000000 1.956
ROI_pairRSPC_anatRV1:trial_typesocial 0.002277 0.057597 149.000000 0.040
ROI_pairRSTS_anatRV1:trial_typesocial -0.113930 0.057597 149.000000 -1.978
Pr(>|t|)
(Intercept) 0.0156 *
ROI_pairRSPC_anatRV1 1.48e-07 ***
ROI_pairRSTS_anatRV1 7.75e-09 ***
trial_typesocial 0.0523 .
ROI_pairRSPC_anatRV1:trial_typesocial 0.9685
ROI_pairRSTS_anatRV1:trial_typesocial 0.0498 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Correlation of Fixed Effects:
(Intr) ROI_pRSPC_RV1 ROI_pRSTS_RV1 trl_ty ROI_RSPC_RV1:
ROI_pRSPC_RV1 -0.451
ROI_pRSTS_RV1 -0.451 0.500
tril_typscl -0.451 0.500 0.500
ROI_RSPC_RV1: 0.319 -0.707 -0.354 -0.707
ROI_RSTS_RV1: 0.319 -0.354 -0.707 -0.707 0.500
ROI_pair = RSPC_RSTS:
contrast estimate SE df t.ratio p.value
physics - social -0.0797 0.0407 149 -1.956 0.0523
ROI_pair = RSPC_anatRV1:
contrast estimate SE df t.ratio p.value
physics - social -0.0819 0.0407 149 -2.012 0.0460
ROI_pair = RSTS_anatRV1:
contrast estimate SE df t.ratio p.value
physics - social 0.0343 0.0407 149 0.841 0.4014
Degrees-of-freedom method: kenward-roger RSPC_RSTS
funcRV1: social–physics difference is greatest for RSPC_RSTS pair than either one with funcRV1. Overall functional connectivity in physics trial (baseline) is higher in RSPC_funcRV1(b = 0.33, p < .001) and RSTS_funcRV1(b = 0.12, p = .002) compared to RSPC_RSTS. Trial_type had a modest effect (b = 0.08, p = .040), meaning there’s a higher connectivity in RSPC_RSTS in social than physical trials. Follow-up comparisons showed that only RSPC_RSTS pair showed significantly higher correlation during social than physical trials (p = .040), but not RSPC_RV1 or RSTS_funcRV1.
anatRV1: RSPC–RSTS connectivity has a marginal increase during social than physical trials. Follow-up comparisons showed that only RSTS–anatRV1 showed significantly higher correlation in social than physical trials, which is contrary to the finding with funcRV1 and our expectation.
interaction(fc_primary_R, roi_set = c("RSMG_RTPJ", "RSMG_funcRV1", "RTPJ_funcRV1"))
interaction(fc_primary_R, roi_set = c("RSMG_RTPJ", "RSMG_anatRV1", "RTPJ_anatRV1"))RSMG_RTPJ
funcRV1: overall functional connectivity in physics trial (baseline) is lower in RSMG_funcRV1(b = -0.15, p < .001) and RTPJ_funcRV1(b = -0.14, p < .001) compared to RSMG_RTPJ. No effect of trial_type.
anatRV1: RSMG_anatRV1 pair showed significantly higher correlation during social than physical trials (b=0.08, p = .02), but not RSMG_TPJ or RTPJ_anatRV1.
interaction(fc_primary_L, roi_set = c("LSPC_LSTS", "LSPC_anatLV1", "LSTS_anatLV1"))
interaction(fc_primary_L, roi_set = c("LSPC_LTPJ", "LSPC_anatLV1", "LTPJ_anatLV1"))LSPC_anatLV1 pair showed significantly higher correlation during social than physical trials, but not LSPC with any other regions.
soc & phys & rest blocks
Visualization
#soc vs. phys vs. rest blocks
fig2_L <- ggplot(fc_primary_L, aes(x = ROI_pair, y=correlation,fill=trial_type))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "grey", "coral2"))+
theme_minimal()+
theme(plot.title = element_text(size=15))+
ylim(-0.5,1)+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
fig2_R <- ggplot(fc_primary_R, aes(x = ROI_pair, y=correlation,fill=trial_type))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "grey", "coral2"))+
theme_minimal()+
theme(plot.title = element_text(size=15))+
ylim(-0.5,1)+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
(fig2_L / fig2_R)
Analysis
How does connectivity in soc vs. phys blocks differ compared to rest, when people are not seeing any stimuli?
we expect to see connectivity increase when people see stimuli, regardless of domain, versus in rest. Do we predict the same for V1?
barrier vs. no barrier
#barrier vs. no-barrier
all_subjects_by_barrier <- all_subjects |>
filter(!is.na(barrier))
fc_primary_L_by_barrier <- get_fc_table(all_subjects_by_barrier, roi_list_primary_L, by_barrier=TRUE)
ggplot(fc_primary_L_by_barrier, aes(x = ROI_pair, y=correlation,fill=barrier))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "coral2"))+
facet_wrap(~trial_type) +
theme_minimal()+
theme(plot.title = element_text(size=15))+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))
fc_primary_R_by_barrier <- get_fc_table(all_subjects_by_barrier,roi_list_primary_R,by_barrier=TRUE)
ggplot(fc_primary_R_by_barrier, aes(x = ROI_pair, y=correlation,fill=trial_type))+
stat_boxplot(geom='errorbar')+
geom_boxplot(position="dodge") +
scale_fill_manual(values=c("deepskyblue", "coral2"))+
facet_wrap(~barrier) +
theme_minimal()+
theme(plot.title = element_text(size=15))+
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust = 1))