Package 'CytoMethIC'

Title: DNA methylation-based machine learning models
Description: This package provides model data and functions for easily using machine learning models that use data from the DNA methylome to classify cancer type and phenotype from a sample. The primary motivation for the development of this package is to abstract away the granular and accessibility-limiting code required to utilize machine learning models in R. Our package provides this abstraction for RandomForest, e1071 Support Vector, Extreme Gradient Boosting, and Tensorflow models. This is paired with an ExperimentHub component, which contains models developed for epigenetic cancer classification and predicting phenotypes. This includes CNS tumor classification, Pan-cancer classification, race prediction, cell of origin classification, and subtype classification models. The package links to our models on ExperimentHub. The package currently supports HM450, EPIC, EPICv2, MSA, and MM285.
Authors: Wanding Zhou [aut] (ORCID: <https://orcid.org/0000-0001-9126-1932>), Jacob Fanale [aut, cre] (ORCID: <https://orcid.org/0009-0002-0490-3269>)
Maintainer: Jacob Fanale <[email protected]>
License: Artistic-2.0
Version: 1.8.0
Built: 2026-10-10 11:18:27 UTC
Source: https://github.com/bioc/CytoMethIC

Help Index

Internal function for fraction optimization

Description

Internal function for fraction optimization

Usage

.optimizeFrac(
  frac,
  ref,
  q,
  errFunc,
  temp = 0.5,
  maxIter = 1000,
  delta = 1e-04,
  step.max = 1,
  verbose = FALSE
)

Arguments

frac

initial fraction
ref

reference
q

query
errFunc

error function
temp

annealing temperature
maxIter

maximum iteration to stop after converge
delta

delta score to reset counter
step.max

maximum step, do not adjust
verbose

output debug info

Value

a list of fractions and min err

Check CytoMethIC versions

Description

print package verison of cytomethic and depended packages to help troubleshoot installation issues.

Usage

cmi_checkVersion()

Value

print the versions of cytomethic and dependencies

Examples

cmi_checkVersion()

Reference-based cell type deconvolution

Description

This is a reference-based cell composition estimation. The function takes a reference methylation status matrix (rows for probes and columns for cell types) and a query beta value measurement.

Usage

cmi_deconvolution(ref, q, trim = FALSE, ...)

Arguments

ref

reference methylation
q

target measurement: length(q) == nrow(ref)
trim

to trim query input beta values. this relieves unclean background subtraction
...

extra parameters for optimization.

Details

The length of the target beta values should be the same as the number of rows of the reference Matrix. The function outputs a list containing the estimated cell fraction, the error of optimization.

Value

a list of fraction, min error.

Examples

ref = cbind(
  CD4 = c(1,1,1,0,1,0),
  CD19 = c(0,0,1,1,0,1),
  CD14 = c(1,1,1,1,0,1))
rownames(ref) = paste0("cg",1:6)
trueFrac = runif(3)
trueFrac = trueFrac / sum(trueFrac)
q = ref %*% trueFrac
trueFrac
cmi_deconvolution(ref, q)

Reference-based cell type deconvolution (allowing one unknown component)

Description

This is a reference-based cell composition estimation. The function takes a reference methylation status matrix (rows for probes and columns for cell types) and a query beta value measurement.

Usage

cmi_deconvolution2(ref, q, trim = FALSE, ...)

Arguments

ref

reference methylation
q

target measurement: length(q) == nrow(ref)
trim

to trim query input beta values. this relieves unclean background subtraction
...

extra parameters to .optimizeFrac

Details

The length of the target beta values should be the same as the number of rows of the reference Matrix. The method assumes one unknown component. It outputs a list containing the estimated cell fraction, the error of optimization and methylation status of the unknown component.

Value

a list of fraction, min error and unknown component methylation state

Examples

ref = cbind(
  CD4 = c(1,1,1,0,1,0),
  CD19 = c(0,0,1,1,0,1),
  CD14 = c(1,1,1,1,0,1))
rownames(ref) = paste0("cg",1:6)
trueFrac = runif(4)
trueFrac = trueFrac / sum(trueFrac)
ref_unk = sample(c(0,1), nrow(ref), replace=TRUE)
q = cbind(ref_unk, ref) %*% trueFrac
trueFrac
res = cmi_deconvolution2(ref, q)
res$frac

Master data frame for all model objects

Description

This is an internal object which will be updated on every new release

Format

tibble

Value

master sheet of CytoMethIC model objects

Examples

print(cmi_models[,c("EHID","Title")])

The cmi_predict function takes in a model and a sample, and uses the model to predict it. This function supports randomForest, e1071::svm, xgboost, and keras/tensorflow models. For xgboost and keras models, the features used in classification as well as a label mapping must be provided for output.

Description

The cmi_predict function takes in a model and a sample, and uses the model to predict it. This function supports randomForest, e1071::svm, xgboost, and keras/tensorflow models. For xgboost and keras models, the features used in classification as well as a label mapping must be provided for output.

Usage

cmi_predict(betas, cmi_model, verbose = FALSE, BPPARAM = SerialParam())

Arguments

betas

DNA methylation beta
cmi_model

Cytomethic model downloaded from ExperimentHub
verbose

be verbose with warning
BPPARAM

use MulticoreParam(n) for parallel processing

Value

predicted cancer type label

Examples

library(sesame)
library(ExperimentHub)
library(CytoMethIC)

## Cancer Type
model = ExperimentHub()[["EH8395"]]
betas = openSesame(sesameDataGet("EPICv2.8.SigDF")[[1]])
betas = imputeBetas(mLiftOver(betas, "HM450"))
cmi_predict(betas, model)

betas = openSesame(sesameDataGet('EPIC.1.SigDF'), mask=FALSE)
cmi_predict(betas, model)

betas = sesameDataGet("HM450.1.TCGA.PAAD")$betas
betas = imputeBetas(betas)
cmi_predict(betas, model)