MethPed (version 1.0.2)

MethPed: MethPed: a DNA methylation classifier tool for the identification of pediatric brain tumor subtypes.

Description

The current version of MethPed can classify the following tumor diagnoses/subgroups: Diffuse Intrinsic Pontine Glioma (DIPG), Ependymoma, Embryonal tumors with multilayered rosettes (ETMR), Glioblastoma (GBM), Medulloblastoma (MB) - Group 3 (MB_Gr3), Group 4 (MB_Gr3), Group WNT (MB_WNT), Group SHH (MB_SHH) and Pilocytic Astrocytoma (PiloAstro).

Usage

MethPed(data, TargetID="TargetID", prob = TRUE,...)

Arguments

data

Data for classification. Data can be in class "ExpressionSet", "matrix" or "data.frame". See MethPed vignette for details.

TargetID

Name of the "Probe" column in data

prob

If 'TRUE' (Default value), the return value will be conditional probability from which tumor group the sample belongs to. 'FALSE' will only return value for the tumor group which has the maximum conditional probability for a sample and other non-maximum values will be zero.

...

More parameter to add

Value

The output of the algorithm is partitioned in 6 parts :

"$target_id"

Probes name in main data set

"$probes"

Number of probes

"$sample"

Number of samples

"$probes_missing"

Names of the probes that were included in the original classifier but are missing from the data at hand.

$oob_err.rate

If a large number of probes are missing this could potentially lead to a drop in the predictive accuracy of the model. Note that, the out-of-bag error is 1.7 percent of the original MethPed classifier.

$predictions

The last and most interesting output contains the predictions. We chose not to assign tumors to one or other group but give probabilities of belonging in one or other subtype. See vignette for details.

Details

Classification of pediatric tumors into biologically defined subtypes is challenging, and multifaceted approaches are needed. For this aim, we developed a diagnostic classifier based on DNA methylation profiles. We offer MethPed as an easy-to-use toolbox that allows researchers and clinical diagnosticians to test single samples as well as large cohorts for subclass prediction of pediatric brain tumors.

The MethPed classifier uses the Random Forest (RF) algorithm to classify unknown pediatric brain tumor samples into sub-types. The classification proceeds with the selection of the beta values needed for the classification.

The computational process proceeded in two stages. The first stage commences with a reduction of the probe pool or building the training probe dataset for classification. We have named this dataset as “predictors”. The second stage is to apply the RF algorithm to classify the probe data of interest based on the training probe dataset (predictors).

For the construction of the training probe pool (predictors), methylation data generated by the Illumina Infinium HumanMethylation 450 BeadChip arrays were downloaded from the Gene Expression Omnibus (GEO). Four hundred seventy-two cases were available, representing several brain tumor diagnoses (DIPG, glioblastoma, ETMR, medulloblastoma, ependymoma, pilocytic astrocytoma) and their further subgroups.

The data sets were merged and probes that did not appear in all data sets were filtered away. In addition, about 190,000 CpGs were removed due to SNPs, repeats and multiple mapping sites. The final data set contained 206,823 unique probes and nine tumor classes including the medulloblastoma subgroups. K–neighbor imputation was used for missing probe data.

After that, a large number of regression analyses were performed to select the 100 probes per tumor class that had the highest predictive power (AUC values). Based on the identified 900 methylation sites, the nine pediatric brain tumor types could be accurately classified using the multiclass classification algorithm MethPed.

References

[1] Anna Danielsson, Szilárd Nemes, Magnus Tisell, Birgitta Lannering, Claes Nordborg, Magnus Sabel, and Helena Carén. "MethPed: A DNA Methylation Classifier Tool for the Identification of Pediatric Brain Tumor Subtypes". Clinical Epigenetics 2015, 7:62, 2015

[2] Breiman, Leo (2001). "Random Forests". Machine Learning 45 (1): 5-32. doi:10.1023/A:1010933404324

[3] Troyanskaya, O., M. Cantor, G. Sherlock, P. Brown, T. Hastie, R. Tibshirani, D. Botstein, and R. B. Altman. "Missing Value Estimation Methods for DNA Microarrays." Bioinformatics 17.6 (2001): 520-25.

See Also

See http://www.clinicalepigeneticsjournal.com/content/7/1/62 for more details.

Examples

#################### Loading and view sample data
data(MethPed_sample)
head(MethPed_sample,10)

#################### Check dimention of sample data
dim(MethPed_sample)  # Check number pof probes and samples in data

#################### Checking missing value in the data
missingIndex <- checkNA(MethPed_sample)

#################### Apply MethPed to sample data (Probability for all tumor group)
myClassification<-MethPed(MethPed_sample)
myClassification<-MethPed(MethPed_sample,prob=TRUE)

#################### Apply MethPed to sample data (Only maximum probability expected)
myClassification_max<-MethPed(MethPed_sample,prob=FALSE)

#################### Summary of results
summary(myClassification)
summary(myClassification)

#################### Barplot of conditional prediction probability on different samples
par(mai = c(1, 1, 1, 2), xpd=TRUE)
mat<-t(myClassification$predictions)
mycols <- c("green",rainbow(nrow(mat),start=0,end=1)[nrow(mat):1],"red")
barplot(mat,  col = mycols,
        beside=FALSE,axisnames=TRUE,
        ylim=c(0,1),xlab= "Sample",ylab="Probability")
legend( ncol(mat)+0.5,1,
       legend = rownames(mat),fill = mycols,xpd=TRUE, cex = 0.6)

## Generic function to plot
plot(myClassification) # myClassification should be an object in "methPed" class