In vjcitn/ccleR6: ccleR6

BiocStyle::markdown()

Sys.setenv("CCLEDB_PATH"="~/BigData/CellLineData/CellLineData.db")
suppressPackageStartupMessages({
library(ccleR6)
})

Introduction

A large indexed SQLite database has been created by Phil Chapman to represent CCLE, Achilles, and other integrative data sources relevant to cancer biology.

This document describes some approaches to user interface design. We indicate how to

• query the database directly
• provide R-level support for substantively useful self-description
• carry out some exemplary 'analyses'

For this code to work you need to have the environment variable CCLEDB_PATH defined to give the path to the SQLite file. ie

Sys.setenv("CCLEDB_PATH"="path/to/db")

An R6 interface

I believe that an object that is somewhat fleshed out relative to the database view provided by dplyr will come in handy. Therefore I defined a reference class and have lightly populated it with some identifier vectors.

ccle = ccledb$new(.ccleSrc)
ccle

This can be serialized, but the database connection needs to be refreshed on load.

The "guide vectors" are created as hints on vocabularies in use.

"BRAF" %in% ccle$cngenes  # copy number gene list

Some guided filters

We have defined a number of filter functions to simplify common subsetting operations.

ccle$src %>% filter_compound("Irinotecan")
ccle$src %>% filter_organ("breast")
ccle$src %>% filter_Histology_patt("%wing%")

Query speed

microbenchmark?

Get Data Function Family

I extended the reference class itself to contain a query set that can be applied across different data sets.

ccle$query_symbols <- c('PTEN', 'BRAF', 'NRAS', 'KRAS', 'SMARCA4', 'PIK3CA')
ccle$query_cell_lines <- (ccle$src %>% filter_organ("skin") %>%  as.data.frame(n=-1))$CCLE_name
ccle$query_compounds <- c('PLX4720', 'AZD6244', 'Lapatinib')

Data can then be retrieved by running a function against the R6 class

affy_data <- get_affy(ccle)
head(affy_data)
hybcap_data <- get_hybcap(ccle)
head(hybcap_data)
resp_data <- get_response(ccle)
head(resp_data)

Importantly, since the data structure is standardised we can now combine disparate data types into a common data frame

combined_data <- bind_rows(affy_data,hybcap_data,resp_data)

Modelling drug response

The get_data functions can be combined to give an output suitable for modeling using the make_df function:

my_df <- make_df(ccle)
my_df[1:5,1:10]
summary(lm(AZD6244_resp ~ as.factor(BRAF_hybcap), data=my_df))
summary(lm(AZD6244_resp ~ as.factor(BRAF_hybcap) + as.factor(NRAS_hybcap), data=my_df))

Visualising drug response

Finally a convenient heatmap plot can be generated using the make_heatmap function to visualise drug response vs genetic features.

make_heatmap(ccle, compound='AZD6244')

Widgets

Two displays of the Barettina paper (Figures 4a [lower dotplot with sensitivity against line] and 4c) can be approached interactively using the multiWidget function. Briefly, multiWidget(ccle) will generate a browser window with two panes, concatenated on one page. The first one allows selection of a gene for which hybrid capture mutation information has been obtained, and a compound. The IC50s for cell lines with hybrid capture data are ordered and displayed. The second panel allows selection of a gene whose expression will be summarized across primary tumor sites.

library(grid)
library(png)
im = readPNG("images/shinyDemo.png")
grid.raster(im)

To do

The vocabularies used for cell lines and tumor anatomy are complicated and need harmonization and streamlining.

Full dose-response information is available and should be exposed.

Interfaces for filters and joins need to be specified and deployed for common use cases.

Additional interactivity such as tooltips over points that can describe mutation profiles.

vjcitn/ccleR6 documentation built on May 3, 2019, 6:14 p.m.