prep_alignment.Rmd

---
title: "Conos alignment of all datasets"
output: html_notebook
date: "`r format(Sys.time(), '%d %B, %Y')`"
---

```{r, message=FALSE, warning=FALSE}
# library(Epilepsy19)
library(dataorganizer)
library(magrittr)
library(Matrix)
library(pbapply)
library(conos)
library(readr)
library(tidyverse)

theme_set(theme_bw())

devtools::load_all()
```

```{r, message=FALSE}
annotation <- read_csv(MetadataPath("annotation.csv"))

cms_all <- CachePath("count_matrices.rds") %>% read_rds()
cms_all <- names(cms_all) %>% setNames(., .) %>% lapply(function(n)
  cms_all[[n]] %>% set_colnames(paste0(n, "_", colnames(.))) %>% 
    set_rownames(make.unique(rownames(.))))

cms_all$NeuN <- NULL
```

## Filter cells

```{r}
cms_filt <- cms_all %>% lapply(function(cm) cm[, (colnames(cm) %in% annotation$cell)])

p2s_filt <- lapply(cms_filt, GetPagoda, graph.k=10, embeding.type="UMAP", n.cores=30,
                   spread=5, min.dist=1.0, build.graph=F)

con_filt <- conos::Conos$new(p2s_filt, n.cores=30)
sample_per_cell <- con_filt$getDatasetPerCell()
condition_per_sample <- ifelse(grepl("ep|HB", levels(sample_per_cell)), "epilepsy", "control") %>%
  setNames(levels(sample_per_cell))

con_filt$buildGraph(k=50, k.self=5, k.self.weight=0.1, same.factor.downweight=0.25, 
                    k.same.factor=5, balancing.factor.per.sample=condition_per_sample)
con_filt$embedGraph(method="UMAP", n.cores=45, min.prob.lower=1e-5, n.neighbors=30, 
                    n.epochs=1000, spread=5, min.dist=1.0, verbose=T)
```

```{r, eval=FALSE}
write_rds(con_filt, CachePath("con_filt_cells.rds"))
```

## Filter samples

```{r, eval=FALSE}
bad_samples <- c("C3", "C5", "E5")

con_sf <- Conos$new(con_filt$samples[!(names(con_filt$samples) %in% bad_samples)], n.cores=30)

con_sf$buildGraph(verbose=T, var.scale=T, k=40, k.self=5, k.self.weight=0.1, 
                  balancing.factor.per.sample=condition_per_sample, k.same.factor=5, 
                  same.factor.downweight=0.25)
con_sf$findCommunities(method=conos::leiden.community, resolution=10)
con_sf$embedGraph(method="UMAP", spread=5, min.dist=1.0, n.epochs=2000, verbose=T, 
                  n.cores=30, min.prob.lower=1e-5)
```

```{r, eval=FALSE}
write_rds(con_sf, CachePath("con_filt_samples.rds"))
```

## Save Pagoda 2 objects

```{r}
# con_sf <- read_rds(CachePath("con_filt_samples.rds"))
# con_filt <- read_rds(CachePath("con_filt_cells.rds"))
```

```{r}
# p2_con <- Pagoda2FromConos(con_sf, n.pcs=0)
# go_env <- pagoda2::p2.generate.human.go(p2_con)
# 
# go_sets <- ExtractGoSets(go_env)
```

### Conos

Filtered samples:

```{r}
# t_anns <- lapply(annotation[2:4], setNames, annotation$cell) %>% 
#   lapply(`[`, names(p2_con$clusters$dataset))
# metadata <- ConvertMetadataToPagoda2Format(
#   AnnotationL1=t_anns$l1, AnnotationL2=t_anns$l2, 
#   AnnotationL3=t_anns$l3, Dataset=p2_con$clusters$dataset
# )
# 
# p2_web <- GetPagodaWebApp(p2_con, con_sf$clusters$leiden$groups, 
#                           additional.metadata=metadata, go.sets=go_sets, go.env=go_env)
# 
# show.app(p2_web, "Samples Filtered")
# p2_web$serializeToStaticFast(CachePath("p2s", "con_filt_samples.bin"));
```

Filtered cells:

```{r}
# con_filt$findCommunities(method=leiden.community, resolution=10, n.iterations=10)
# p2_con_filt <- Pagoda2FromConos(con_filt, n.pcs=0)
# t_anns <- lapply(annotation[2:4], setNames, annotation$cell) %>% lapply(`[`, names(p2_con_filt$clusters$dataset))
# metadata <- ConvertMetadataToPagoda2Format(AnnotationL1=t_anns$l1, AnnotationL2=t_anns$l2,
#                                            AnnotationL3=t_anns$l3, Dataset=p2_con_filt$clusters$dataset)
# 
# p2_web_filt <- GetPagodaWebApp(p2_con_filt, con_filt$clusters$leiden$groups, additional.metadata=metadata,
#                                go.sets=go_sets, go.env=go_env)
# 
# # show.app(p2_web_filt, "Cells Filtered")
# p2_web_filt$serializeToStaticFast(CachePath("p2s", "con_filt_cells.bin"));
```

### Individual

```{r}
# p2 <- con_filt$samples[[1]]
# p2s_web <- lapply(con_filt$samples, function(p2) {
#   p2$makeKnnGraph(k=30, type="PCA", center=T, distance="cosine", weight.type="none", verbose=F)
#   p2$getKnnClusters(type="PCA", method=leiden.community, resolution=7, n.iterations=10, name="leiden")
# 
#   t_anns <- lapply(annotation[2:4], setNames, annotation$cell) %>% lapply(`[`, names(p2$clusters$PCA$leiden))
#   metadata <- ConvertMetadataToPagoda2Format(AnnotationL1=t_anns$l1, AnnotationL2=t_anns$l2, AnnotationL3=t_anns$l3)
#   GetPagodaWebApp(p2, p2$clusters$PCA$leiden, additional.metadata=metadata, go.sets=go_sets, go.env=go_env)
# })
```

```{r}
# for (n in names(p2s_web)) {
#   p2s_web[[n]]$serializeToStaticFast(CachePath("p2s", paste0(n, ".bin")));
# }
```