README.md

In cgeisenberger/crystalmeth: Streamlined Methylation-Based Classification of Tumors

Introduction

Installation

To install crystalmeth from within R, copy and run the following code:

install.packages("devtools")
devtools::install_github("cgeisenberger/crystalmeth")

Motivation

DNA methylation is increasingly being used as a diagnostic tool. This is facilitated by array-based platforms such as Illuminas 450k array and its successor, the EPIC platform. The research community has produced a number of classifiers which are able to determine tissue of origin and / or tumor subclass solely based on DNA methylation data as input.

Capper et al. provide compelling evidence that array-based DNA methylation profiling of tumors can augment (or arguably even replace) microscopy-based histopathology. This has led to the recommendation by the WHO to perform methylation-based profiling for a number of well-characterized brain tumor entities.

It seems highly likely that this approach which has been spearheaded by the neuropathological community will soon be used in other tumor entities. This has prompted us to develop a software package which allows users to perform tasks related to methylation-based classification in a coherent framework that is scalable and can be deployed as a web application.

Primer on DNA Methylation Data

Crystalmeth is designed to work with array-based methylation data. Raw data is obtained by measuring fluorescence for two different dyes on the physical slides in Illumina scanners. Raw data therefore consists of two matched files which end with either _Red.idat or _Grn.idat. The preceding characters correspond to the sample ID, which is also referred to as basename (example: 201533570046_R01C01_Grn.idat).

In addition to the biological signal, multiple technical sources can add noise to the measurements. Normalization can mitigate some of these effects. For a given cytosine base, methylation is usually expressed as a percentage. This measure is also referred to as beta value. While DNA methylation is binary state on the single-nucleotide level, arrays measure the average across all different cell populations present in the (bulk) input material. Beta values can therefore take any value between 0 and 1 (however, samples show a clear bimodal distribution with most probes being either or fully methylated or unmethylated).

Since the details around the analysis of array-based DNA methylation data is beyond the scope of this vignette, we would like to point readers towards the minfi R package developed by Kasper Daniel Hansen. It provides more information about the fundamentals of DNA methylation data analysis and is also used as the backbone for sample preprocessing in this software package.

How to use the package

Example Data

Data to reproduce the example below can be downloaded via this Dropbox link.

Example Workflow

# load necessary libraries
library(tidyverse)
library(crystalmeth)

# load data (see section 'Example Data' above)
rf_model <- readRDS(file = "./testing/test-data/rf_model.RDS")
ridge_model <- readRDS(file = "./testing/test-data/calibration_model.RDS")

# detect input data (IDAT files)
input_dir <- "./testing/test-data/"
idats <- scan_directory(dir = input_dir)
example <- idats %>% get_cases()

# process case
case <- ClassificationCase$new(basename = example,
                               path = input_dir,
                               verbose = TRUE)
case$run_full_workflow(rf_model = rf_model,
                       calibration_model = ridge_model)

# render report
render_report(case = case,
              input = "./testing/test-data/nen_id_template_html.Rmd",
              output_file = paste0(case$array_basename, ".html"),
              output_dir = "./testing/test-data/")

The report created for this sample can be viewed here

More information

For a more thorough walkthrough, consult the package vignette.

cgeisenberger/crystalmeth documentation built on Aug. 28, 2020, 9:35 a.m.