ngram_analysis.Rmd

---
title: "ngram analysis"
author: "Katarzyna Sidorczuk"
date: '2022-06-14'
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = FALSE, fig.width = 14, fig.height = 10, warning = FALSE, message = FALSE)
library(dplyr)
library(biogram)
library(targets)
library(ranger)
library(ggplot2)
library(DT)
library(stringr)
library(ggbeeswarm)
library(patchwork)

source("./functions/ngram_model_functions.R")
source("./functions/ensemble_model_functions.R")

tar_load(c(PlastoGram_best_architecture, envelope_traintest_ngram_matrix, envelope_traintest_data_df,
           ngram_matrix_traintest_graphpart, data_df_graphpart))

my_DT <- function(df, title) {
  DT::datatable(df, extensions = 'FixedColumns', filter = "top", style = "bootstrap",
                options = list(dom = 't', scrollX = TRUE, scrollCollapse = TRUE,
                               scrollY = "700px", paging = FALSE, server = FALSE),
                caption = htmltools::tags$caption(style = 'caption-side: top; text-align: center; color:black; font-size:150% ;', title) )
}

ihs <- function(x) {
  y <- log(x + sqrt(x ^ 2 + 1))
  return(y)
}

train_rf <- function(ngram_matrix, target, imp_ngrams, with_class_weights = FALSE, smote = FALSE) {
  ranger_train_data <- if(smote == TRUE) {
    do_smote(ngram_matrix[, imp_ngrams], target, setNames(as.list(unname(1/(table(target)/max(table(target))))), names(table(target)))) %>% 
      mutate(tar = as.factor(tar))
  } else {
    data.frame(ngram_matrix[, imp_ngrams],
               tar = as.factor(target))
  }
  
  if(with_class_weights == FALSE) {
    class_weights <- NULL
  } else {
    class_weights <- sapply(levels(ranger_train_data[["tar"]]), function(i) 1/sum(ranger_train_data[["tar"]] == i)/nrow(ranger_train_data), USE.NAMES = FALSE)
  }
  ranger(dependent.variable.name = "tar", data =  ranger_train_data, 
         write.forest = TRUE, probability = TRUE, num.trees = 500, 
         verbose = FALSE, class.weights = class_weights,
         seed = 427244, importance = "impurity")
}

new_PlastoGram_ngram_models <- train_ngram_models(PlastoGram_best_architecture, envelope_traintest_ngram_matrix, envelope_traintest_data_df, filtering_colname = NULL, filtering_term = NULL)
new_PlastoGram_ngram_models_graphpart <- train_ngram_models(PlastoGram_best_architecture, ngram_matrix_traintest_graphpart, data_df_graphpart, filtering_colname = NULL, filtering_term = NULL)

```

# Holdout version

## Serine distribution in features of Nuclear_model

```{r}
imp <- new_PlastoGram_ngram_models[["Nuclear_model"]][["variable.importance"]]
serine_frac <- sapply(names(imp), function(i) str_count(i, "S")/nchar(i))
plot_dat <- data.frame(ngram = names(imp),
                       feature_importance = imp,
                       serine_fraction = serine_frac)

pp <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = feature_importance)) +
  geom_jitter(size = 0.5, height = 0.2) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pp_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = ihs(feature_importance))) +
  geom_jitter(size = 0.5, height = 0.2) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pq <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = feature_importance)) +
  geom_quasirandom(size = 0.5, groupOnX = FALSE, width = 0.3) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pq_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = ihs(feature_importance))) +
  geom_quasirandom(size = 0.5, groupOnX = FALSE, width = 0.3) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pd <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(x = feature_importance)) +
  geom_density() +
  theme_bw() +
  scale_y_reverse()

pd_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(x = ihs(feature_importance))) +
  geom_density() +
  theme_bw() +
  scale_y_reverse()

pp/pd + plot_layout(heights = c(3, 1))
pq/pd + plot_layout(heights = c(3, 1))
pp_ihs/pd_ihs + plot_layout(heights = c(3, 1))
pq_ihs/pd_ihs + plot_layout(heights = c(3, 1))
```

## L18 motif ngrams in Membrane model
```{r}
imp_m <- new_PlastoGram_ngram_models[["Membrane_model"]][["variable.importance"]]
motifs <- sapply(names(imp_m), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18 <- "VDPLYPGGSFDPLGLADD"
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_m),
                       feature_importance = imp_m,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_m)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  theme_bw()

```

## L18 motif ngrams in Nuclear membrane model
```{r}
imp_nm <- new_PlastoGram_ngram_models[["Nuclear_membrane_model"]][["variable.importance"]]
motifs <- sapply(names(imp_nm), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_nm),
                       feature_importance = imp_nm,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_nm)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  theme_bw()

p <- plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 50)  %>% 
  mutate(ngram = sapply(.[["ngram"]], function(i) {
    ifelse(nchar(i) == 1, i, paste0(unlist(strsplit(decode_ngrams(i), "")), collapse = " "))
  })) %>%
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  xlab("Gini importance") +
  theme_bw(base_size = 8) +
  scale_fill_manual(values = c("grey70", "#76d872"))

p

ggsave("DPLG_motif_holdout.eps", p, width = 5, height = 5,
       path = "~/Dropbox/Projekty/BioNgramProjects/PlastoGram/Publication_results/")
```
## L18 motif ngrams in N_E vs. N_TM model

```{r}
imp_tm <- new_PlastoGram_ngram_models[["N_E_vs_N_TM_model"]][["variable.importance"]]
motifs <- sapply(names(imp_tm), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_tm),
                       feature_importance = imp_tm,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_tm)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()
```



## Features with the highest Gini importance

```{r}
#save(new_PlastoGram_ngram_models, file = "new_PlastoGram_ngram_models.rda")
#load("new_PlastoGram_ngram_models.rda")
lapply(names(new_PlastoGram_ngram_models), function(ith_model) {
  x <- importance(new_PlastoGram_ngram_models[[ith_model]]) %>% 
    as.data.frame() %>% 
    tibble::rownames_to_column() %>% 
    setNames(c("ngram", "Gini importance")) %>% 
    arrange(`Gini importance`) %>% 
    mutate(ngram = factor(ngram, levels = .[["ngram"]]))
  
  tail(x, n = 100) %>% 
    ggplot(aes(x = ngram, y = `Gini importance`)) +
    geom_col() + 
    coord_flip() +
    theme_bw() +
    ggtitle(paste0(ith_model, " - 100 most important of all ", nrow(x), " informative features"))
})
```

## All features with Gini importance > 0

```{r}
importance(new_PlastoGram_ngram_models[["Nuclear_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>%
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Nuclear_model")

importance(new_PlastoGram_ngram_models[["Membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Membrane_model")

importance(new_PlastoGram_ngram_models[["Plastid_membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Plastid_membrane_model")

importance(new_PlastoGram_ngram_models[["Nuclear_membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Nuclear_membrane_model")

importance(new_PlastoGram_ngram_models[["N_E_vs_N_TM_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("N_E_vs_N_TM_model")

importance(new_PlastoGram_ngram_models[["N_E_vs_N_S_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("N_E_vs_N_S_model")
```


# Partitioning version

## Serine distribution in features of Nuclear_model

```{r}
imp <- new_PlastoGram_ngram_models_graphpart[["Nuclear_model"]][["variable.importance"]]
serine_frac <- sapply(names(imp), function(i) str_count(i, "S")/nchar(i))
plot_dat <- data.frame(ngram = names(imp),
                       feature_importance = imp,
                       serine_fraction = serine_frac)

pp <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = feature_importance)) +
  geom_jitter(size = 0.5, height = 0.2) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pp_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = ihs(feature_importance))) +
  geom_jitter(size = 0.5, height = 0.2) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pq <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = feature_importance)) +
  geom_quasirandom(size = 0.5, groupOnX = FALSE, width = 0.3) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pq_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(y = serine_fraction, x = ihs(feature_importance))) +
  geom_quasirandom(size = 0.5, groupOnX = FALSE, width = 0.3) + 
  theme_bw() + 
  theme(axis.text.x = element_blank(), axis.ticks.x = element_blank(), axis.title.x = element_blank())

pd <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(x = feature_importance)) +
  geom_density() +
  theme_bw() +
  scale_y_reverse()

pd_ihs <- plot_dat %>% 
  mutate(serine_fraction = as.factor(round(serine_fraction, 2))) %>% 
  ggplot(., aes(x = ihs(feature_importance))) +
  geom_density() +
  theme_bw() +
  scale_y_reverse()

pp/pd + plot_layout(heights = c(3, 1))
pq/pd + plot_layout(heights = c(3, 1))
pp_ihs/pd_ihs + plot_layout(heights = c(3, 1))
pq_ihs/pd_ihs + plot_layout(heights = c(3, 1))
```

## L18 motif ngrams in Membrane model
```{r}
imp_m <- new_PlastoGram_ngram_models_graphpart[["Membrane_model"]][["variable.importance"]]
motifs <- sapply(names(imp_m), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18 <- "VDPLYPGGSFDPLGLADD"
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_m),
                       feature_importance = imp_m,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_m)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  theme_bw()

```

## L18 motif ngrams in Nuclear membrane model
```{r}
imp_nm <- new_PlastoGram_ngram_models_graphpart[["Nuclear_membrane_model"]][["variable.importance"]]
motifs <- sapply(names(imp_nm), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_nm),
                       feature_importance = imp_nm,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_nm)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  theme_bw()

p <- plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 50)  %>% 
  mutate(ngram = sapply(.[["ngram"]], function(i) {
    ifelse(nchar(i) == 1, i, paste0(unlist(strsplit(decode_ngrams(i), "")), collapse = " "))
  })) %>%
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = DPLG)) +
  geom_col() +
  xlab("Gini importance") +
  theme_bw(base_size = 8) +
  scale_fill_manual(values = c("grey70", "#76d872"))

p

ggsave("DPLG_motif_partitioning.eps", p, width = 5, height = 5,
       path = "~/Dropbox/Projekty/BioNgramProjects/PlastoGram/Publication_results/")
```
## L18 motif ngrams in N_E vs. N_TM model

```{r}
imp_tm <- new_PlastoGram_ngram_models_graphpart[["N_E_vs_N_TM_model"]][["variable.importance"]]
motifs <- sapply(names(imp_tm), function(i) ifelse(nchar(i) != 1, gsub("_", ".", decode_ngrams(i)), i), USE.NAMES = FALSE)
l18_motifs <- sapply(motifs, function(i) grepl(i, l18), USE.NAMES = FALSE)
dplg_motifs <- sapply(motifs, function(i) grepl(i, "DPLG"), USE.NAMES = FALSE)
plot_dat <- data.frame(ngram = names(imp_tm),
                       feature_importance = imp_tm,
                       L18 = l18_motifs,
                       DPLG = dplg_motifs)

print(paste0("Number of all features: ", length(imp_tm)))
print(paste0("Number of features contained within L18 motif: ", sum(l18_motifs)))
print(paste0("Number of features contained within DPLG motif: ", sum(dplg_motifs)))

ggplot(plot_dat, aes(x = L18, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

ggplot(plot_dat, aes(x = DPLG, y = feature_importance)) +
  geom_quasirandom() +
  theme_bw() 

plot_dat %>% 
  arrange(desc(feature_importance)) %>% 
  head(n = 100) %>% 
  mutate(ngram = factor(ngram, levels = rev(.[["ngram"]]))) %>% 
  ggplot(aes(y = ngram, x = feature_importance, fill = L18)) +
  geom_col() +
  theme_bw()
```



## Features with the highest Gini importance

```{r}
#save(new_PlastoGram_ngram_models, file = "new_PlastoGram_ngram_models.rda")
#load("new_PlastoGram_ngram_models.rda")
lapply(names(new_PlastoGram_ngram_models_graphpart), function(ith_model) {
  x <- importance(new_PlastoGram_ngram_models_graphpart[[ith_model]]) %>% 
    as.data.frame() %>% 
    tibble::rownames_to_column() %>% 
    setNames(c("ngram", "Gini importance")) %>% 
    arrange(`Gini importance`) %>% 
    mutate(ngram = factor(ngram, levels = .[["ngram"]]))
  
  tail(x, n = 100) %>% 
    ggplot(aes(x = ngram, y = `Gini importance`)) +
    geom_col() + 
    coord_flip() +
    theme_bw() +
    ggtitle(paste0(ith_model, " - 100 most important of all ", nrow(x), " informative features"))
})
```

## All features with Gini importance > 0

```{r}
importance(new_PlastoGram_ngram_models_graphpart[["Nuclear_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>%
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Nuclear_model")

importance(new_PlastoGram_ngram_models_graphpart[["Membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Membrane_model")

importance(new_PlastoGram_ngram_models_graphpart[["Plastid_membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Plastid_membrane_model")

importance(new_PlastoGram_ngram_models_graphpart[["Nuclear_membrane_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("Nuclear_membrane_model")

importance(new_PlastoGram_ngram_models_graphpart[["N_E_vs_N_TM_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("N_E_vs_N_TM_model")

importance(new_PlastoGram_ngram_models_graphpart[["N_E_vs_N_S_model"]]) %>% 
  as.data.frame() %>% 
  tibble::rownames_to_column() %>% 
  setNames(c("ngram", "Gini importance")) %>% 
  filter(`Gini importance` > 0) %>% 
  arrange(desc(`Gini importance`)) %>% 
  my_DT("N_E_vs_N_S_model")
```