Ordination.rmd

---
title: "Ordination Lan"
author: "Lan Huong Nguyen"
date: "9/7/2018"
output: 
  BiocStyle::html_document:
    toc_float: true
    df_print: paged
---

```{r setup, include=FALSE}
rm(list = ls())
knitr::opts_chunk$set(
  echo = TRUE, warning = FALSE, message = FALSE, 
  fig.path = "../figs/ordination/", dev=c('png', 'pdf'),
  fig.align = 'center', fig.wide = TRUE, 
  fig.width = 8, fig.height = 6
)
```

# Set up

```{r}
.packages <- c("gridExtra","RColorBrewer", "viridis", "phyloseq", "tidyverse")
sapply(.packages, require, character.only = TRUE)

options(stringsAsFactors = FALSE)
theme_set(theme_bw())
theme_update(
  text = element_text(size = 20),
  strip.background = element_rect(fill = "grey95"))
```

```{r}
collapse_samples <- function(physeq, sample_id = "SampleID", fun = mean){
  sampledata_collapsed <- sample_data(physeq) %>%
    select(-Lane, -MeasID, -OrigSampleDepth) %>%
    distinct() 
  rownames(sampledata_collapsed) <- sampledata_collapsed[[sample_id]]
  seqtab <- data.frame(as(otu_table(physeq), "matrix"))
  if(!taxa_are_rows(physeq)) seqtab <- t(seqtab)
  
  sample_ids <- physeq@sam_data[[sample_id]]
  seqtab_collapsed <- sapply(unique(sample_ids), function(id){ 
    if(sum(sample_ids == id) == 1) return(seqtab[, sample_ids == id])
    apply(seqtab[, sample_ids == id], 1, fun)
  })
  rownames(seqtab_collapsed) <- rownames(seqtab)
  physeq_collapsed <- phyloseq(
    otu_table(seqtab_collapsed, taxa_are_rows = TRUE),
    sample_data(sampledata_collapsed), 
    tax_table(physeq), phy_tree(physeq))
  return(physeq_collapsed)
}


get_scores <- function(pcoa_out, smp_data, axes = 1:2){
  scores <- pcoa_out$vectors[, axes]
  colnames(scores) <- paste0("PC", axes)
  # Combine sample scores and sample data
  sample_scores <- data.frame(scores, stringsAsFactors = FALSE) %>%
    rownames_to_column("SampleID") %>%
    left_join(smp_data) %>%
    mutate(
      SubjectID_Timepoint = factor(paste0(SubjectID, "_", Timepoint)),
      Timepoint = factor(Timepoint, levels = c("Baseline", "10 Week"))
    )
  return(sample_scores)
}

get_evals <- function(pcoa_out) {
  evals <- pcoa_out$values[,1]
  var_exp <- 100 * evals/sum(evals)
  return(list("evals" = evals, "variance_exp" = var_exp))
}


plot_ordination <- function(
  sample_scores, var_exp, group_column = "SubjectID_Timepoint", 
  color_continuous = FALSE, color_column = NULL, colors = NULL, shape = 21, 
  ax_names = c("PC1", "PC2")){
  if(is.null(color_column)){
    color_column <- "Color_Column"
    sample_scores$Color_Column <- "Color"
  }
  if(is.null(colors)){
    nCols <-length(unique(sample_scores[[color_column]]))
    colors <- RColorBrewer::brewer.pal(9, "Set1")
    colors <- colorRampPalette(colors)(nCols)
  }
  plt <- ggplot(
    sample_scores, 
    aes_string(
      x = ax_names[1], y = ax_names[2], fill = color_column)) +
    geom_polygon(aes_string(group = group_column)) +
    geom_path(aes_string(group = group_column, color = color_column)) +
    geom_point(size = 3, color = "grey10", shape = shape) +
    xlab(sprintf("PC1 [%s%% variance]", round(var_exp[1], 2))) +
    ylab(sprintf("PC2 [%s%% variance]", round(var_exp[2], 2))) +
    coord_fixed() +
    theme(text = element_text(size = 20)) 
  if (!color_continuous){ # if continuous
    plt <- plt + scale_fill_manual(values = colors, na.value = "grey") +
    scale_color_manual(values = colors, na.value = "grey") 
  }
  return(plt)
}
```

# Load data

```{r}
(ps0 <- readRDS("../data/phyloseq/diet_participants_tipglom_with_adherence.rds"))
```

## Filter out samples

```{r}
minSampleDepth <- 10000
(ps0 <- subset_samples(ps0, sample_sums(ps0) > minSampleDepth))
# Outlier samples:
(ps0 <- subset_samples(ps0, !(SubjectID == "1139" & Timepoint == "10 Week")))
(ps0 <- subset_samples(ps0, !is.na(Cohort)))
(ps0 <- subset_samples(ps0, !is.na(Cat.12Mo)))
(ps0 <- subset_taxa(ps0, taxa_sums(ps0) > 0))

sample_data(ps0)$Subj_Group <- 
  factor(paste0(sample_data(ps0)$SubjectID, "_", sample_data(ps0)$Timepoint))
```

## Data Stats

```{r}
table(ps0@sam_data$Cohort)
```

```{r}
summary(sample_sums(subset_samples(ps0, Cohort == 2)))
summary(sample_sums(subset_samples(ps0, Cohort == 3)))
```


```{r}
SUBJ <- sample_data(ps0) %>%
  data.frame() %>%
  select(SubjectID, Cat.12Mo, Cohort, Diet, Gender, Timepoint,
         Weight_kg.BL, Age, Percent.12Mo) %>%
  distinct() 

df <- SUBJ %>%
  group_by(Cohort, Diet, Timepoint) %>%
  summarise(Count = n()) %>%
  arrange(Cohort,Timepoint, Diet)
df
```

```{r}
df <- SUBJ %>%
  group_by(Cohort, Diet, Timepoint, Gender) %>%
  summarise(Count = n()) %>%
  arrange(Cohort,Timepoint, Diet, Gender)
df
```

```{r}
SUBJ <- sample_data(ps0) %>%
  filter(Timepoint == "Baseline") %>%
  data.frame() %>%
  select(SubjectID, Cat.12Mo, Cohort, Diet, Gender, 
         Weight_kg.BL, Age, Percent.12Mo) %>%
  distinct() 

SUBJ %>%
  group_by(Cohort, Diet) %>%
  summarise(count = n())
```

```{r}
SUBJ %>%
  group_by(Cohort, Diet, Gender) %>%
  summarise(count = n())
```

```{r}
SUBJ <- sample_data(ps0) %>%
  filter(Timepoint == "Baseline") %>%
  data.frame() %>%
  select(SubjectID, Cat.12Mo, Cohort, Diet, Gender, 
         Weight_kg.BL, Weight_kg.12Mo, Age, Percent.12Mo) %>%
  distinct() 

SUBJ %>% 
  group_by(Cohort, Diet) %>%
  summarise(
    m_age = mean(Age), sd_age = sd(Age),
    m_weight = mean(Weight_kg.BL), sd_wBL = sd(Weight_kg.BL), 
    m_w12Mo = mean(Weight_kg.12Mo), sd_w12Mo = sd(Weight_kg.12Mo), 
    m_ploss= mean(Percent.12Mo), sd_ploss = sd(Percent.12Mo))
```

```{r}
SUBJ %>% group_by(Cohort, Cat.12Mo) %>%
  summarise(count = n()) %>%
  arrange(Cohort, Cat.12Mo)
```

Biological samples which were sequenced twice:

```{r}
(sample_data(ps0) %>% 
   group_by(Cohort, SampleID) %>% 
   dplyr::summarise(
     count = n(), 
     meas = paste0(MeasID, collapse = "; ")
     ) %>%
   filter(count > 1))
```

```{r}
new_smp_data <- sample_data(ps0) %>%
  data.frame() %>%
  select(-Lane, -MeasID, -OrigSampleDepth) %>%
  distinct() 
rownames(new_smp_data) <- new_smp_data$SampleID

dim(new_smp_data)
```

Combine repeated samples by computing the mean 

```{r}
ps <- collapse_samples(ps0)
(ps <- subset_taxa(ps, taxa_sums(ps) > 0))
#sample_data(ps)$Cat.12Mo[is.na(sample_data(ps)$Cat.12Mo)] <- "Unknown"
# sample_data(ps)$Cat.12Mo <- 
#   factor(sample_data(ps)$Cat.12Mo, levels = c("US", "MS", "VS", "Unknown")) 
sample_data(ps)$Cat.12Mo <-
  factor(sample_data(ps)$Cat.12Mo, levels = c("US", "MS", "VS"))
```


## Experimental Design

```{r}
diet_cols <- c("US" = "#E41A1C", "MS" = "#377EB8", "VS" = "#4DAF4A", "Unknown" = "grey")
```


```{r exp-design-cohort2-low-carb, fig.height=6.5, fig.width=10}
subj_bl_weight <- sample_data(ps) %>%
  select(SubjectID, Weight_kg.BL) %>%
  distinct() %>%
  arrange(Weight_kg.BL)
SMP <- data.frame(sample_data(ps)) %>%
  mutate(
    SubjectID_factor = factor(SubjectID, levels = subj_bl_weight$SubjectID),
    Cohort = ifelse(Cohort == 2, "Discovery", "Validation"))
  
plt_exp_carb <- ggplot(
  SMP %>% filter(Diet == "Low-carb", Cohort == "Discovery"),
  aes(y = SubjectID_factor, x = SampleDate)) + 
  geom_point(aes(shape = Timepoint), fill = diet_cols[1], size = 4) +
  scale_shape_manual(values = c("Baseline" = 21, "10 Week" = 24)) +
  ylab("SubjectID") + xlab("Sample Data")
plt_exp_carb
```


```{r exp-design-cohort2-low-fat, fig.height=6.5, fig.width=10}
plt_exp_purple <- ggplot(
  SMP %>% filter(Diet == "Low-fat", Cohort == "Discovery"),
  aes(y = SubjectID_factor, x = SampleDate)) + 
  geom_point(aes(shape = Timepoint), size = 4, fill = diet_cols[2]) +
  scale_shape_manual(values = c("Baseline" = 21, "10 Week" = 24)) +
  ylab("SubjectID") + xlab("Sample Data")
plt_exp_purple
```

```{r exp-design-cohort3-low-carb, fig.height=5, fig.width=7}
plt_exp_carb <- ggplot(
  SMP %>% filter(Diet == "Low-carb", Cohort == "Validation"),
  aes(y = SubjectID_factor, x = SampleDate)) + 
  geom_point(aes(shape = Timepoint), fill = diet_cols[1], size = 4) +
  scale_shape_manual(values = c("Baseline" = 21, "10 Week" = 24)) +
  ylab("SubjectID") + xlab("Sample Data")
plt_exp_carb
```

```{r exp-design-cohort3-low-fat, fig.height=5, fig.width=7}
plt_exp_purple <- ggplot(
  SMP %>% filter(Diet == "Low-fat", Cohort == "Validation"),
  aes(y = SubjectID_factor, x = SampleDate)) + 
  geom_point(aes(shape = Timepoint), size = 4, fill = diet_cols[2]) +
  scale_shape_manual(values = c("Baseline" = 21, "10 Week" = 24)) +
  ylab("SubjectID") + xlab("Sample Data")
plt_exp_purple
```




## Data transformation

```{r}
psasinh <- transform_sample_counts(ps, function(x) {asinh(x)})
psasinh_cohort2 <- subset_samples(psasinh, Cohort == 2)
(psasinh_cohort2 <- subset_taxa(psasinh_cohort2, taxa_sums(psasinh_cohort2) > 0))
psasinh_cohort3 <- subset_samples(psasinh, Cohort == 3)
(psasinh_cohort3 <- subset_taxa(psasinh_cohort3, taxa_sums(psasinh_cohort3) > 0))
```




# Princopal Coordinate Analysis (PCoA) on asinh data

```{r}
## Ordinate with Bray-Curtis distance
pcoa_all <- ordinate(psasinh,  method = "MDS", distance = "bray")
pcoa_cohort2 <- ordinate(psasinh_cohort2,  method = "MDS", distance = "bray")
pcoa_cohort3 <- ordinate(psasinh_cohort3,  method = "MDS", distance = "bray")

# Compute variance explained
var_exp_all <- get_evals(pcoa_all)$variance_exp
var_exp_cohort2 <- get_evals(pcoa_cohort2)$variance_exp
var_exp_cohort3 <- get_evals(pcoa_cohort3)$variance_exp

scores_all <- get_scores(pcoa_all, sample_data(psasinh))
scores_cohort2 <- get_scores(pcoa_cohort2, sample_data(psasinh_cohort2))
scores_cohort3 <- get_scores(pcoa_cohort3, sample_data(psasinh_cohort3))
```


## Weight-loss Success Categories

* PCoA on all samples for both cohorts:

```{r pcoa-bray-all, fig.height=8, fig.width=15}
tmp_shapes <- ifelse(scores_all %>% filter(!is.na(Diet), !is.na(Cohort)) %>%
                       .[["Timepoint"]] == "Baseline", 21, 24)
  
plot_ordination(scores_all %>% filter(!is.na(Diet), !is.na(Cohort)),
  var_exp_all, color_column = "Cat.12Mo", 
  colors = diet_cols, shape = tmp_shapes) +
  facet_grid(Timepoint ~ Cohort + Diet, drop = TRUE) +
  guides(
    fill = guide_legend(
      title = "Weight Loss Success", 
      override.aes = list(shape = 21)),
    color = "none")
```

* PCoA on all samples for both cohorts only baseline samples


```{r pcoa-bray-all-bl, fig.height=5, fig.width=12}
plot_ordination(
  scores_all %>% 
    filter(!is.na(Diet), !is.na(Cohort), Timepoint == "Baseline"),
  var_exp_all, color_column = "Cat.12Mo", colors = diet_cols) +
  facet_grid(~  Diet ) +
   guides(
    fill = guide_legend(
      title = "Weight Loss Success", 
      override.aes = list(shape = 21)),
    color = "none")
```

* PCoA on all samples for both cohorts showing only baseline samples and
cohorts separately


```{r pcoa-bray-all-bl-facet, fig.height=10, fig.width=12}
plot_ordination(
  scores_all %>% 
    filter(!is.na(Diet), !is.na(Cohort), Timepoint == "Baseline"),
  var_exp_all, color_column = "Cat.12Mo", colors = diet_cols) +
  facet_grid(Cohort ~ Diet) +
   guides(
    fill = guide_legend(
      title = "Weight Loss Success", 
      override.aes = list(shape = 21)),
    color = "none")
```


* PCoA on samples from cohort 2


```{r pcoa-bray-cohort2, fig.height=10, fig.width=12}
tmp_shapes <- ifelse(
  scores_cohort2 %>% filter(!is.na(Diet)) %>% .[["Timepoint"]] == "Baseline", 
  21, 24)
   
plot_ordination(scores_cohort2 %>% filter(!is.na(Diet)),
  var_exp_cohort2,color_column = "Cat.12Mo", 
  colors = diet_cols, shape = tmp_shapes) +
  facet_grid(Timepoint ~ Diet) +
   guides(
    fill = guide_legend(
      title = "Weight Loss Success", 
      override.aes = list(shape = 21)),
    color = "none")
```

* PCoA on samples from cohort 2 only baseline samples shown

```{r pcoa-bray-cohort2-bl, fig.height=6, fig.width=12}
discov_bl <- scores_cohort2 %>% 
  filter(!is.na(Diet), Timepoint == "Baseline") %>%
  mutate(Cohort = "Discovery")

plot_ordination(discov_bl,
  var_exp_cohort2,color_column = "Cat.12Mo", colors = diet_cols) +
  facet_grid( ~ Diet, drop = TRUE) +
   guides(
    fill = guide_legend(
      title = "Weight Loss \n Success", 
      override.aes = list(shape = 21)),
    color = "none")
```


```{r  pcoa-bray-cohort2-bl-convex-hull, fig.height=6, fig.width=12}
chulls <- plyr::ddply(
  discov_bl, c("Cat.12Mo", "Diet"), 
  function(df) df[chull(df$PC1, df$PC2), ])


ggplot(discov_bl, aes(x = PC1, y = PC2, fill = Cat.12Mo)) +
  geom_polygon(data = chulls,aes(group = Cat.12Mo), alpha = 0.1) +
  geom_polygon(aes(group = SubjectID_Timepoint)) +
  geom_path(aes(group = SubjectID_Timepoint, color = Cat.12Mo)) +
  geom_point(size = 3, color = "grey10", shape = 21) +
  xlab(sprintf("PC1 [%s%% variance]", round(var_exp_cohort2[1], 2))) +
  ylab(sprintf("PC2 [%s%% variance]", round(var_exp_cohort2[2], 2))) +
  coord_fixed() + theme(text = element_text(size = 20)) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_fill_manual(values = diet_cols, na.value = "grey") +
  scale_color_manual(values = diet_cols, na.value = "grey") +
  guides(
    fill = guide_legend(
      title = "Weight Loss Success", 
      override.aes = list(shape = 21)),
    color = "none") +
  theme(legend.position = "top", text = element_text(size = 25))


```


* PCoA on samples from cohort 3 (only baseline samples avaiable)

```{r pcoa-bray-cohort3, fig.height=5, fig.width=12}
valid <- scores_cohort3 %>% 
  filter(!is.na(Diet)) %>%
  mutate(Cohort = "Validation")

plot_ordination(valid,
  var_exp_cohort3,color_column = "Cat.12Mo", colors = diet_cols) +
  facet_grid( ~ Diet) +
   guides(
    fill = guide_legend(
      title = "Weight Loss \n Success", 
      override.aes = list(shape = 21)),
    color = "none")
```

```{r  pcoa-bray-cohort3-bl-convex-hull, fig.height=6, fig.width=12}

chulls <- plyr::ddply(
  valid, c("Cat.12Mo", "Diet"),
  function(df) df[chull(df$PC1, df$PC2), ])


ggplot(valid, aes(x = PC1, y = PC2, fill = Cat.12Mo)) +
  geom_polygon(data = chulls, aes(group = Cat.12Mo), alpha = 0.1) +
  geom_polygon(aes(group = SubjectID_Timepoint)) +
  geom_path(aes(group = SubjectID_Timepoint, color = Cat.12Mo)) +
  geom_point(size = 3, color = "grey10", shape = 21) +
  xlab(sprintf("PC1 [%s%% variance]", round(var_exp_cohort3[1], 2))) +
  ylab(sprintf("PC2 [%s%% variance]", round(var_exp_cohort3[2], 2))) +
  coord_fixed() + theme(text = element_text(size = 20)) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_fill_manual(values = diet_cols, na.value = "grey") +
  scale_color_manual(values = diet_cols, na.value = "grey") +
 guides(
  fill = guide_legend(
    title = "Weight Loss Success", 
    override.aes = list(shape = 21)),
  color = "none") +
  theme(legend.position = "top", text = element_text(size = 25))

```


```{r pcoa-bray-all-bl-convex-hull, fig.height=10, fig.width=12}
discov_bl <- discov_bl %>%
  left_join(ps0@sam_data[, c("SampleID", "Lane")])
valid <- valid %>%
  left_join(ps0@sam_data[, c("SampleID", "Lane")])
bl_DF <- bind_rows(discov_bl, valid)
chulls <- plyr::ddply(bl_DF, c("Cat.12Mo", "Diet", "Cohort"),
                      function(df) df[chull(df$PC1, df$PC2), ])


ggplot(bl_DF, aes(x = PC1, y = PC2, fill = Cat.12Mo)) +
  geom_polygon(data = chulls, aes(group = Cat.12Mo), alpha = 0.2) +
  geom_polygon(aes(group = SubjectID_Timepoint)) +
  geom_path(aes(group = SubjectID_Timepoint, color = Cat.12Mo)) +
  geom_point(size = 3, color = "grey10", shape = 21) +
  xlab(sprintf("PC1 [%s%% variance]", round(var_exp_cohort2[1], 2))) +
  ylab(sprintf("PC2 [%s%% variance]", round(var_exp_cohort2[2], 2))) +
  coord_fixed() + theme(text = element_text(size = 20)) +
  facet_grid(Cohort ~ Diet, drop = TRUE) +
  scale_fill_manual(values = diet_cols, na.value = "grey") +
  scale_color_manual(values = diet_cols, na.value = "grey")

```


## BMI

* PCoA on samples from cohort 2 only baseline samples shown


```{r pcoa-bray-all-bl-bmi, fig.height=5, fig.width=16}

plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "bmi_bl", color_continuous = TRUE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()

```


```{r pcoa-bray-cohort2-bl-bmi, fig.height=6, fig.width=12}
plot_ordination(discov_bl ,
  var_exp_cohort2, color_column = "bmi_bl", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE)  +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-bmi, fig.height=5, fig.width=12}
plot_ordination(valid %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "bmi_bl", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```

## Mean Adherence

```{r}
bl_DF <- bl_DF %>%
  mutate_at(.vars = vars(contains("adherence_")), .funs = as.numeric) %>%
  mutate(
    mean_adherence = rowMeans(select(., adherence_12_months:adherence_6_months)),
    mean_reduction = adherence_baseline - mean_adherence,
    mean_adherence = 100 - mean_adherence,
  )


discov_bl <- discov_bl %>%
  mutate_at(.vars = vars(contains("adherence_")), .funs = as.numeric) %>%
  mutate(
    mean_adherence = rowMeans(select(., adherence_12_months:adherence_6_months)),
    mean_reduction = adherence_baseline - mean_adherence,
    mean_adherence = 100 - mean_adherence,
  )

valid <- valid  %>%
  mutate_at(.vars = vars(contains("adherence_")), .funs = as.numeric) %>%
  mutate(
    mean_adherence = rowMeans(select(., adherence_12_months:adherence_6_months)),
    mean_reduction = adherence_baseline - mean_adherence,
    mean_adherence = 100 - mean_adherence,
  )
```


* PCoA on samples from cohort 2 only baseline samples shown


```{r pcoa-bray-all-bl-adherence, fig.height=5, fig.width=16}

plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "mean_adherence", color_continuous = TRUE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()

```


```{r pcoa-bray-cohort2-bl-adherence, fig.height=6, fig.width=12}
plot_ordination(discov_bl ,
  var_exp_cohort2, color_column = "mean_adherence", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE)  +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-adherence, fig.height=5, fig.width=12}
plot_ordination(valid %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "mean_adherence", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```


## Mean reduction/diet fidelity

* PCoA on samples from cohort 2 only baseline samples shown


```{r pcoa-bray-all-bl-reduction, fig.height=5, fig.width=16}

plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "mean_reduction", color_continuous = TRUE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()

```


```{r pcoa-bray-cohort2-bl-reduction, fig.height=6, fig.width=12}
plot_ordination(discov_bl ,
  var_exp_cohort2, color_column = "mean_reduction", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE)  +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-reduction, fig.height=5, fig.width=12}
plot_ordination(valid %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "mean_reduction", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_color_viridis_c() +
  scale_fill_viridis_c()
```


## Sequencing Run

* PCoA on samples from cohort 2 only baseline samples shown


```{r pcoa-bray-all-bl-seqrun, fig.height=5, fig.width=16}

plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "Lane", color_continuous = TRUE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE) 

```


```{r pcoa-bray-cohort2-bl-seqrun, fig.height=6, fig.width=12}
plot_ordination(discov_bl ,
  var_exp_cohort2, color_column = "Lane", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) 
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-seqrun, fig.height=5, fig.width=12}
plot_ordination(valid %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "Lane", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) 
```


## Subject Age

* PCoA on samples from cohort 2 only baseline samples shown




```{r pcoa-bray-all-bl-age, fig.height=5, fig.width=16}
plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "Age", color_continuous = TRUE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE) +
  scale_color_viridis(option = 3, end = 0.90) +
  scale_fill_viridis(option = 3, end = 0.90) 
```


```{r pcoa-bray-cohort2-bl-age, fig.height=6, fig.width=12}

plot_ordination(discov_bl,
  var_exp_cohort2,color_column = "Age", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_color_viridis(option = 3, end = 0.90) +
  scale_fill_viridis(option = 3, end = 0.90) 
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-age, fig.height=5, fig.width=12}
plot_ordination(scores_cohort3 %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "Age", color_continuous = TRUE) +
  facet_grid( ~ Diet, drop = TRUE) +
  scale_color_viridis(option = 3, end = 0.9) +
  scale_fill_viridis(option = 3, end = 0.9) 
```


## Gender

* PCoA on samples from cohort 2 only baseline samples shown

```{r pcoa-bray-all-bl-gender, fig.height=5, fig.width=16}
plot_ordination(bl_DF,
  var_exp_cohort2,color_column = "Gender", colors = c("#E41A1C", "#377EB8"), 
  color_continuous = FALSE) +
  facet_grid( ~ Cohort + Diet, drop = TRUE)
```

```{r pcoa-bray-cohort2-bl-gender, fig.height=6, fig.width=12}
plot_ordination(scores_cohort2 %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort2,color_column = "Gender", colors = c("#E41A1C", "#377EB8"), 
  color_continuous = FALSE) +
  facet_grid( ~ Diet, drop = TRUE)
```

* PCoA on samples from cohort 3 only baseline samples shown


```{r pcoa-bray-cohort3-bl-gender, fig.height=5, fig.width=12}
plot_ordination(scores_cohort3 %>% filter(!is.na(Diet), Timepoint == "Baseline"),
  var_exp_cohort3, color_column = "Gender", colors = c("#E41A1C", "#377EB8"), 
  color_continuous = FALSE) + facet_grid( ~ Diet, drop = TRUE)
```

## Subject Shift over time

* PCoA on samples from cohort 2

```{r pcoa-bray-cohort2-timeshift, fig.height=12, fig.width=10}

DF <- scores_cohort2 %>%
  group_by(SubjectID, Cat.12Mo, Diet, Timepoint) %>%
  summarise(PC1 = mean(PC1), PC2 = mean(PC2)) %>%
  filter(!is.na(Diet))
  
ggplot(DF, aes(x = PC1, y = PC2)) +
  geom_path(aes(group = SubjectID), lwd = 1.5, color = "grey50") +
  geom_text(aes(color = Timepoint, label = SubjectID), size = 4, fontface = "bold") +
  facet_grid(Cat.12Mo ~ Diet) + 
  scale_color_manual(
    values = c("Baseline" = brewer.pal(8, "Dark2")[5], 
               "10 Week" = brewer.pal(8, "Dark2")[4])) +
  xlab(sprintf("PC1 [%s%% variance]", round(var_exp_cohort2[1], 2))) +
  ylab(sprintf("PC2 [%s%% variance]", round(var_exp_cohort2[2], 2))) +
  coord_fixed() + theme(text = element_text(size = 15)) 

```








# PERMANOVA test for overall difference between weight-loss success groups

(Permutational multivariate analysis of variance)
Null hypothesis: there is no difference in between weight-loss success groups

* Bray-Curtis distance on sequence-level data


```{r}
smpdata <- sample_data(ps0) %>%
  data.frame() %>%
  select(MeasID, SubjectID, Timepoint)

subjdata <- sample_data(ps0) %>%
  data.frame() %>%
  select(Cohort, Diet, SubjectID, Gender, Age, Weight_kg.BL, Weight_kg.12Mo,
         Cat.12Mo, Percent.12Mo, Timepoint) %>%
  distinct() %>%
  mutate(SubjectID_Timepoint = paste0(SubjectID, "_", Timepoint)) %>%
  column_to_rownames("SubjectID_Timepoint")

seqtab <- t(asinh(as(otu_table(ps0), "matrix")))
seqtab_subj <- aggregate(
  seqtab, FUN = mean,
  by = list(smpdata[rownames(seqtab), "SubjectID"], 
            smpdata[rownames(seqtab), "Timepoint"])) %>%
  mutate(SubjectID_Timepoint = paste0(Group.1, "_", Group.2)) %>%
  column_to_rownames("SubjectID_Timepoint")

seqtab_subj_mat <- as.matrix(seqtab_subj %>% select(-contains("Group.")))
seqtab_subj_mat <- t(seqtab_subj_mat)

psasinh_subj <- phyloseq(otu_table(seqtab_subj_mat, taxa_are_rows = TRUE),
                         sample_data(subjdata))
```


```{r}
psasinh_subj_bl <- subset_samples(psasinh_subj, Timepoint == "Baseline")

adonis_res <- list()
for(cohort in c(2, 3)) {
  cohort_name <- ifelse(cohort == 2, "Discovery", "Validation")
  cat("Testing for cohort:", cohort_name, ".\n")
  for(diet in c("Low-carb", "Low-fat")){
    cat("Testing for diet:", diet, ".\n")
    test_name <- paste0("Cohort", ".", cohort_name, "_", "Diet", ".", diet)
    ips <- psasinh_subj_bl %>%
        subset_samples(Diet == diet & Cohort == cohort)
    D <- phyloseq::distance(ips, method = "bray")
    adonis_res[[test_name]] <- 
      vegan::adonis(D ~ Cat.12Mo, data = data.frame(sample_data(ips)))
    
  }
}

```


```{r}
adonis_res[["Cohort.Discovery_Diet.Low-carb"]]
```


```{r}
adonis_res[["Cohort.Discovery_Diet.Low-fat"]]
```


```{r}
adonis_res[["Cohort.Validation_Diet.Low-carb"]]
```


```{r}
adonis_res[["Cohort.Validation_Diet.Low-fat"]]
```


```{r}
# psasinh.bySubject <- collapse_physeq_bySubject(psasinh)
# bl.psasinh.blue <- subset_samples(psasinh.bySubject, Timepoint == "Baseline" & Color == "Blue" & Cohort == 2) 
# metadata.blue <- as(bl.psasinh.blue@sam_data, "data.frame")
# set.seed(325525)
# adonis(distance(bl.psasinh.blue, method = "bray") ~ Cat.12Mo, data = metadata.blue)
# 
# bl.psasinh.purple <- subset_samples(psasinh.bySubject, Timepoint == "Baseline" & Color == "Purple" & Cohort == 2)
# metadata.purple <- as(bl.psasinh.purple@sam_data, "data.frame")
# set.seed(325525)
# purp.adonis <- adonis(distance(bl.psasinh.purple, method = "bray") ~ Cat.12Mo, data = metadata.purple)
# purp.adonis
```