In drorberel/Bioc2mlr:

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>", message=FALSE)

library(tidyverse)  
library(magrittr)

library(Bioc2mlr)
library(mlr)
select<-dplyr::select

Model-evaluation (ML)

A. SummarizedExperiment (SE)

Convert raw data from SE S4 class, to mlr's "task"

data(Golub_Merge, package = 'golubEsets') # ExpressionSet 
smallG<-Golub_Merge[200:259,]
smallG

library(SummarizedExperiment)
smallG_SE<-makeSummarizedExperimentFromExpressionSet(smallG)

# functional:
task_SE_Functional<-Fun_SE_to_taskFunc(smallG_SE, param.Y.name = 'ALL.AML', param.covariates = NULL, param_positive_y_level = 'ALL', task_return_format = 'functional', task_type = 'classif') ## will work with either 1 or multiple assayS
task_SE_Functional


# non-functional:
## 1. directly, but into DF
extracted_DF_from_task_SE<-getTaskData(task_SE_Functional, functionals.as = "dfcols") # keep matrix
extracted_DF_from_task_SE[,1:10] %>% str


## 2. Fun_SE_to_taskFunc(..., task_return_format = 'dfcols')
task_SE_NON_Functional<-Fun_SE_to_taskFunc(smallG_SE, param.Y.name = 'ALL.AML', param.covariates = NULL, param_positive_y_level = 'ALL', task_return_format = 'dfcols', task_type = 'classif') ## will work with either 1 or multiple assayS

## 3. functional_to_NonFunctional_task_function(task_functional)
task_SE_NON_Functional_alt<-functional_to_NonFunctional_task_function(task_SE_Functional)



## 4. designated function ## TBA
# extracted = extractFDAFeatures(task_SE_Functional, feat.methods = list("exprs" = all))

Single assay ML demonstration

Direct

library(class)
smallG_train<-exprs(smallG)[,1:40]     %>% t 
smallG_test <-exprs(smallG)[,-c(1:40)] %>% t
knn_pred<-knn(smallG_train, smallG_test, cl = smallG$ALL.AML[1:40], k = 1, prob=TRUE)
table(smallG$ALL.AML[-c(1:40)], knn_pred)

MLInterface

library(MLInterfaces)
krun<-MLearn(formula = ALL.AML~., data = smallG, .method = knnI(k=1), trainInd = 1:40)
krun
confuMat(krun)

mlr

task_train<-task_SE_Functional %>% subsetTask(subset = 1:40)
task_test <-task_SE_Functional %>% subsetTask(subset = 41:72)
classif.lrn = makeLearner("classif.knn")
model<-train(classif.lrn, task_train)
Predict<-model %>% predict(task_test)
Predict %>% calculateConfusionMatrix()

B. MultiAssayExperiment (MAE)

Two data examples:
1. miniACC, balanced, without 'dropouts'.
2. Customized, non-balanced, with 'dropouts'.

Convert raw data from MAE S4 class, to mlr's "task"

1. miniACC

library(MultiAssayExperiment)
miniACC
# miniACC %>% sampleMap %>% data.frame %>% dplyr::select(primary, assay) %>% table # no replicates within same assay

task_Functional_MAE<-Fun_MAE_to_taskFunc(miniACC, param.Y.name = 'vital_status', param.covariates = c('gender','days_to_death'), param_positive_y_level = '1', task_type = 'classif')
task_Functional_MAE
extracted_DF_from_task_MAE_functionals<-getTaskData(task_Functional_MAE, functionals.as = "matrix") # keep functionals
extracted_DF_from_task_MAE_functionals[,1:10] %>% glimpse

extracted_DF_from_task_MAE_dfcols<-getTaskData(task_Functional_MAE, functionals.as = "dfcols") # concatonate functionals
extracted_DF_from_task_MAE_dfcols[,1:10] %>% glimpse

2. Customized

library(MultiAssayExperiment)

patient.data <- data.frame(sex=c("M", "F", "M", "F", "F"),
                           age=38:42,
                           row.names=c("Jack", "Jill", "Bob", "Barbara","Meg"))
## assay A
arraydat <- matrix(seq(101, 108), ncol=4,
                    dimnames=list(c("ENST00000294241", "ENST00000355076"),
                                  c("array1", "array2", "array3", "array4")))
coldat <- data.frame(slope53=rnorm(4), row.names=c("array1", "array2", "array3", "array4"))
exprdat <- SummarizedExperiment(arraydat, colData=coldat)
exprmap <- data.frame(primary=c("Jill", "Jill", "Meg", "Barbara"),
                       colname=c("array1", "array2", "array3", "array4"),
                       stringsAsFactors = FALSE)
## assay B
methyldat <-
    matrix(1:10, ncol=5,
           dimnames=list(c("ENST00000355076", "ENST00000383706"),
                         c("methyl1", "methyl2", "methyl3",
                           "methyl4", "methyl5")))
methylmap <- data.frame(primary = c("Jack", "Jack", "Jack", "Meg", "Bob"),
                         colname = c("methyl1", "methyl2", "methyl3", "methyl4", "methyl5"),
                         stringsAsFactors = FALSE)

myMultiAssay <- MultiAssayExperiment(list("A" = exprdat, "B" = methyldat), patient.data, list(A = exprmap, B = methylmap) %>% listToMap)
myMultiAssay
myMultiAssay %>% sampleMap %>% data.frame %>% select(primary, assay) %>% table # Yes replicates within same assay, and non-balanced  / dropouts!!!

# myMultiAssay %>% sampleMap %>% data.frame %>% filter(assay == 'A')
# myMultiAssay$sex

task_Functional_MAE_customized<-Fun_MAE_to_taskFunc(myMultiAssay, param.Y.name = 'sex', param.covariates = NULL, param_positive_y_level = 'M', task_type = 'classif')

Multi-assay ML demonstration

mlr: vertical integration

Unless the learner has sepecific implementation for functional data, it will be automatically converted into standard (non-functional) task.
bartMachine model was chosed only because it has a built-in NA handling. Any other 'learner' from mlr could be demonstrated instead.

library(bartMachine)
classif_lrn_bartMachine<-makeLearner("classif.bartMachine")
model_bartMachine<-train(classif_lrn_bartMachine, task_Functional_MAE)
Predict_bartMachine<-model_bartMachine %>% predict(task_Functional_MAE)
Predict_bartMachine %>% calculateConfusionMatrix()

TBA: built-in CPOs / see detailed vignette.

Session information