In ComputationalIntelligenceGroup/gardenr: Easily access the gardener's classification labels for interneurons

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

gardenr

The goal of gardenr is provide simple functions of analyzing the data from Gardener's interneuron classification as well as to provide examples of analyses of such data.

Installation

You can use the devtools package to install gardenr from Github.

devtools::install_github('ComputationalIntelligenceGroup/gardenr')

The user needs to provide the path to the folder containing the data. These can be downloaded from https://figshare.com/s/8a761698160a675bb080.

folder <- '/home/bmihaljevic/code-gardener/gardener-experiment-data/data/' 
library(tidyverse)

Usage

gardenr functions return the data for all 320 cells. There are functions for the classification labels get_all_labels, metadata get_all_meta, and alternative type names and definitions get_alternative_types.

library(gardenr)
annotations <- get_all_labels(folder ) 
summary(annotations)

alternative <- get_alternative_types(folder)
summary(alternative)

metadata <- get_all_meta(folder)
summary(metadata)

To get data on a specific cell or annotator, we only need filter by id:

metadata %>% filter(neuron == 12)
head(annotations %>% filter(neuron == 12))

head(annotations %>% filter(annotator == 2))
alternative %>% filter(annotator == 2)

It is easy to combine the three data frames using joins, and then compute summaries with tidyverse functions. For example, the alternative type definition to an annotation

head(left_join(annotations, alternative, by = c("other" = "type", "annotator" = "annotator"))) 

The primary key in alternative consits of annotator and type and hence we need both in the join.

There is a utility function to return the counts for all categories, with a single entry per neuron.

counts <- get_all_counts(folder) 
head(counts)

We can join the above with metadata to get thorough information for each cell.

meta_count <- left_join(counts, metadata, by = 'neuron') %>% select(-paper, -original.type, -archive, -neuromorpho.name)
head(meta_count)

There is a utility function doing just that:

head(get_all_counts_meta(folder) )

These data are also available in a different format, which is a bit easier to analyze

annot_tidy <- make_annotations_tidy(annotations) 
head(annot_tidy )

Examples

• How many partially labelled cells?

table(annotations$complete) 

• Get raw frequencies of the different categories

annot_tidy <- make_annotations_tidy(annotations) 
annot_tidy %>% group_by(feature, value) %>% tally()
# library(ggplot2)
ggplot(annot_tidy, aes(x = value, color = feature)) + geom_bar() + theme(axis.text.x = element_text(angle = 90, hjust = 1))

• Only consider cells with above > 30 neurocientists agreeing on a specific label

annot_tidy <- make_annotations_tidy(annotations) 
above30 <- annot_tidy %>% group_by(neuron, feature, value) %>% tally() %>% filter(n > 30)
above30 %>% group_by(feature) %>% summarize(n_distinct(neuron))  

• Restrict analysis to unrotated, monkey cells

monkey <- metadata %>% filter(species == 'Monkey')
counts <- get_all_counts_meta(folder)
counts %>% filter(species == 'Monkey') 

Alternative type names

• 27 out of 48 neuroscientists provided alternative type names for the cells

length(unique(alternative$annotator))
unique(alternative$annotator) 

• How many alternative types per neuroscientist? How many other cells per neuroscientist?

types <- alternative %>% group_by(annotator) %>% tally()
other <- annotations %>% group_by(annotator) %>% filter(F5 == 'other') %>% tally()
ggplot(data.frame(types = types$n, cells = other$n), aes(x = types, y = cells)) + geom_point()

• Words used in alternative types and definitions:
  • The most common word is `bitufted'

library(tidytext)
library(wordcloud)
alt <- alternative 
alt$type <- as.character(alternative$type ) 
alt <- alt %>% unnest_tokens(type, type)
alt %>%  count(type, sort = TRUE)      
alt %>%  count(type, sort = TRUE)      %>%  with(wordcloud(type, n, random.order = FALSE, max.words = 50 , colors= brewer.pal(8,"Dark2")))

alt <- alternative 
alt$definition <- as.character(alternative$definition )   
alt <- alt %>% unnest_tokens(definition, definition)
alt <- alt %>% anti_join(get_stopwords(), by = c("definition" = "word"))
alt %>%  count(definition, sort = TRUE)      
alt %>%  count(definition, sort = TRUE)      %>%  with(wordcloud(definition, n, random.order = FALSE, max.words = 50 , colors= brewer.pal(8,"Dark2")))

ComputationalIntelligenceGroup/gardenr documentation built on Sept. 5, 2019, 12:58 p.m.