within_clust_sort: within_clust_sort

In jrboyd/seqsetvis: Set Based Visualizations for Next-Gen Sequencing Data

within_clust_sort

Description

Without modifying cluster assignments, modify the order of rows within each cluster based on within_order_strategy.

Usage

within_clust_sort(
  clust_dt,
  row_ = "id",
  column_ = "x",
  fill_ = "y",
  facet_ = "sample",
  cluster_ = "cluster_id",
  within_order_strategy = c("hclust", "sort", "left", "right", "none", "reverse")[2],
  clustering_col_min = -Inf,
  clustering_col_max = Inf,
  dcast_fill = NA
)

Arguments

clust_dt	data.table output from ssvSignalClustering
row_	variable name mapped to row, likely id or gene name for ngs data. Default is "id" and works with ssvFetch* output.
column_	varaible mapped to column, likely bp position for ngs data. Default is "x" and works with ssvFetch* output.
fill_	numeric variable to map to fill. Default is "y" and works with ssvFetch* output.
facet_	variable name to facet horizontally by. Default is "sample" and works with ssvFetch* output. Set to "" if data is not facetted.
cluster_	variable name to use for cluster info. Default is "cluster_id".
within_order_strategy	one of "hclust", "sort", "right", "left", "reverse". If "hclust", hierarchical clustering will be used. If "sort", a simple decreasing sort of rosSums. If "left", will atttempt to put high signal on left ("right" is opposite). If "reverse" reverses existing order (should only be used after meaningful order imposed).
clustering_col_min	numeric minimum for col range considered when clustering, default in -Inf
clustering_col_max	numeric maximum for col range considered when clustering, default in Inf
dcast_fill	value to supply to dcast fill argument. default is NA.

Details

This is particularly useful when you want to sort within each cluster by a different variable from cluster assignment. Also if you've imported cluster assigments but want to sort within each for the new data for a prettier heatmap.

TODO refactor shared code with clusteringKmeansNestedHclust

Value

data.table matching input clust_dt save for the reassignment of levels of row_ variable.

Examples

data(CTCF_in_10a_profiles_dt)
#clustering by relative value per region does a good job highlighting changes
#when then plotting raw values the order within clusters is not smooth
#this is a good situation to apply a separate sort within clusters.
prof_dt = CTCF_in_10a_profiles_dt
prof_dt = append_ynorm(prof_dt)
prof_dt[, y_relative := y_norm / max(y_norm), list(id)]

clust_dt = ssvSignalClustering(prof_dt, fill_ = "y_relative")
clust_dt.sort = within_clust_sort(clust_dt)

cowplot::plot_grid(
  ssvSignalHeatmap(clust_dt) +
    labs(title = "clustered by relative, sorted by relative"),
  ssvSignalHeatmap(clust_dt.sort) +
    labs(title = "clustered by relative, sorted by raw value")
)

jrboyd/seqsetvis documentation built on Dec. 10, 2024, 11:23 a.m.