LipidFrag

MetFrag based lipid identification

Requirements

• Java >= 1.6
• R >= 3.1.1
• R packages: gdata, ROCR, jsonlite

MetFrag

As input LipidFrag requires a MetFrag generated candidate list by using a LipidMaps database. LipidMaps identifiers encode the lipid ontology which is used by LipidFrag. When using MetFragCLI or MetFragR LipidMaps can be included by downloading the database file from the MSBI website and setting the database:

settingsObject[["MetFragDatabaseType"]]<-"LocalCSV"
settingsObject[["LocalDatabasePath"]]<-"PATH_TO_LIPIDMAPS_CSV"

or for MetFragCLI

MetFragDatabaseType=LocalCSV
LocalDatabasePath=PATH_TO_LIPIDMAPS_CSV

The generated MetFrag (CSV) output file can then be used as input for

predict.lipidmaps.class

function. See lipidfrag_main.r to try an example.

R scripts

lipidfrag_train.r

• includes functions to train prediction models based on lipid standard data
• uses files model_scores_pos.txt and model_scores_neg.txt for prediction models in positive and negative mode

models.pos <- generate.model("../data/model_scores_pos.txt")
models.neg <- generate.model("../data/model_scores_neg.txt")

lipidfrag_predict.r

• includes functions to predict lipid classes based on trained models and MetFrag CSV files generated by using LipidMaps as candidate source

lipidfrag_functions.r

• includes general functions used by lipidfrag_train.r and lipidfrag_predict.r

lipidfrag_main.r

• includes an example showing how to use LipidFrag
• necessary data files are included in the repository

Additionals

Plot model value distribution

• after training the models score values can be plotted together with trained distribution parameters

# distributions for negative ionization models
sapply(names(models.neg), function(class) {plot.model.data(models.neg[[class]], main = paste("Distribution of ", class, " (neg)", sep=""))})
# distributions for positive ionization models
sapply(names(models.pos), function(class) {plot.model.data(models.pos[[class]], main = paste("Distribution of ", class, " (pos)", sep=""))})

Plot ROC curves

• after training the models ROC curves can be plotted to get quality of classifications based on given standard data

# ROC curves for negative ionization models
sapply(names(models.neg), function(class) calculate.roc("../data/model_scores_pos.txt", paste(class, models.pos[[class]]$type, sep="_"), main = paste("ROC of ", class, " (pos)", sep=""), variant = 1))
# ROC curves for positive ionization models
sapply(names(models.pos), function(class) calculate.roc("../data/model_scores_neg.txt", paste(class, models.neg[[class]]$type, sep="_"), main = paste("ROC of ", class, " (neg)", sep=""), variant = 1))