Add tSNE plot of null simulations

analysis/fdr.Rmd

@@ -45,9 +45,9 @@ fdr_data <- fdr_pars %>%
   unchop(c(sel_mgs, group_id))
 ```
 
-```{r count-signficant-genes}
+```{r n-sig-true-null}
 fdr_data %>%
-  filter(sim_label == "standard_sim") %>% 
+  filter(sim_label == "null_sim") %>% 
   filter(is.na(test.use) | test.use != "roc") %>% 
   filter(is.na(pval.type) | pval.type == "all") %>% 
   rowwise() %>% 
@@ -55,43 +55,66 @@ fdr_data %>%
   ungroup() %>% 
   group_by(pars, rep) %>% 
   summarise(n_sig = sum(n_sig), .groups = "drop") %>% 
-   mutate(pars = fct_reorder(factor(pars), n_sig)) %>% 
+  group_by(pars) %>% 
+  summarise(n_sig = mean(n_sig), .groups = "drop") %>%
+  mutate(pars = fct_reorder(factor(pars), n_sig)) %>% 
   ggplot(aes(pars, y = n_sig)) + 
-  geom_boxplot() + 
-  coord_flip() + 
+  geom_col() + 
+  geom_text(
+    aes(label = n_sig),
+    position = position_dodge(0.1),
+    hjust = -0.5, size = 3,
+  ) + 
+  coord_flip(ylim = c(0, 11000)) + 
+  scale_y_continuous(breaks = seq(0, 10000, by = 2000)) + 
   labs(
+    title = "Null simulations",
     x = "Method", 
     y = "Number of significant genes"
   ) + 
   theme_bw()
 ```
 
-```{p-value-histograms}
-fdr_data %>% 
-  filter(sim_label == "standard_sim" & rep == 1) %>% 
+```{r n-sig-standard}
+fdr_data %>%
+  filter(sim_label == "standard_sim") %>% 
   filter(is.na(test.use) | test.use != "roc") %>% 
+  filter(is.na(pval.type) | pval.type == "all") %>% 
   rowwise() %>% 
-  mutate(p_value = list(sel_mgs$p_value)) %>% 
+  mutate(n_sig = filter(sel_mgs, p_value_adj < 0.05) %>% nrow()) %>% 
   ungroup() %>% 
-  unnest(cols = p_value) %>% 
-  ggplot(aes(x = p_value)) + 
-  geom_histogram(bins = 30) + 
-  facet_wrap(~pars) + 
+  group_by(pars, rep) %>% 
+  summarise(n_sig = sum(n_sig), .groups = "drop") %>% 
+  group_by(pars) %>% 
+  summarise(n_sig = mean(n_sig), .groups = "drop") %>%
+  mutate(pars = fct_reorder(factor(pars), n_sig)) %>% 
+  ggplot(aes(pars, y = n_sig)) + 
+  geom_col() + 
+  geom_text(
+    aes(label = n_sig), 
+    position = position_dodge(0.1),
+    hjust = -0.5, size = 3,
+  ) + 
+  coord_flip(ylim = c(0, 24000)) + 
+  scale_y_continuous(breaks = seq(0, 24000, by = 4000)) + 
   labs(
-    x = "p value", 
-    y = "Count"
+    title = "Standard simulations", 
+    x = "Method", 
+    y = "Number of significant genes"
   ) + 
-  theme_bw() 
+  theme_bw()
+```
 
+```{r p-value-histograms}
+# With knowledge of the true null.
 fdr_data %>% 
-  filter(sim_label == "standard_sim" & rep == 1) %>% 
+  filter(sim_label == "null_sim" & rep == 1 ) %>% 
   filter(is.na(test.use) | test.use != "roc") %>% 
   rowwise() %>% 
-  mutate(p_value = list(sel_mgs$p_value)) %>% 
+  mutate(p_value_adj = list(sel_mgs$p_value_adj)) %>% 
   ungroup() %>% 
-  unnest(cols = p_value) %>% 
-  mutate(p_value = log(p_value + 0.001)) %>% 
-  ggplot(aes(x = p_value)) + 
+  unnest(cols = p_value_adj) %>% 
+  ggplot(aes(x = p_value_adj)) + 
   geom_histogram(bins = 30) + 
   facet_wrap(~pars) + 
   labs(