## Data consist of 150 samples of organic molecular mixtures analyzed by
## pyrolysis gas chromatography-mass spectrometry (py-GC-MS).
## Each data set has the scan numbers in the rows and the mass to charge ratio (m/z)
## values in the columns.
## The data values are the raw intensities as measured by py-GC-MS.
## Each sample has either an abiotic (A) or biotic (B) origin.
## Purpose is to determine classification rules to predict whether the samples are
## biotic or abiotic.
## Purpose is also to determine the features, m/z and scan numbers, that discriminate
## between the biotic and the abiotic groups.
## Logistic regression with elastic net penalty is used as the machine learning method.
## We apply nested resampling to obtain the generalized prediction performance.
## The data are preprocessed before we use machine learning.
library(dplyr) # for dataframe computation
##
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
##
## filter, lag
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
library(MALDIquant) # for chemometrics processing
##
## This is MALDIquant version 1.22.2
## Quantitative Analysis of Mass Spectrometry Data
## See '?MALDIquant' for more information about this package.
library(caret) # for machine learning
## Loading required package: ggplot2
## Loading required package: lattice
library(mlr3) # for machine learning
library("mlr3verse") # for machine learning
library("mlr3learners")# for machine learning
library("mlr3tuning") # for machine learning
## Loading required package: paradox
library("data.table") # for rbindlist
##
## Attaching package: 'data.table'
## The following objects are masked from 'package:dplyr':
##
## between, first, last
library(glmnet) # for logistic regression with elastic net penalty
## Loading required package: Matrix
## Loaded glmnet 4.1-8
#Data preparation
setwd("C:/Desktop/Templeton_grant_research/Second_paper_material/Data2022")
species=c(rep("A",5),"B", "A","A","B","B","A","C", rep("B",5),rep("C",4),rep("B",4),"A",
rep("B",11),"A","A","B","A","A","B","B","A",rep("B",3),"A","B",rep("A",6),
rep("B",3),"A","A","B","B",rep("A",4),"B",rep("A",5),"B",rep("A",10),
rep("B",4),rep("A",5),rep("B",3),rep("A",3),rep("B",4),rep("A",8),"B",
"A","A","B","C","C","A",rep("B",3),"A","B", "B","A","B",rep("A",3),"B","A",
rep("A",7),"B","A","B")
species2=as.numeric(as.factor(species))
ind=which(species2=="3")
species_n=species2[-ind]
#Reading in the data is based on the code found in:
## "How to import multiple .csv files simultaneously in R and create a data frame"
## by Dahna, A., datascience+, 03/09/2019
#https://datascienceplus.com/how-to-import-multiple-csv-files-simultaneously-in-r-and-create-a-data-frame/
species_names <- list.files()
species_names=species_names[-ind]
z=lapply(species_names, read.delim) #read in all of 134 datasets
setwd("C:/Desktop/Templeton_grant_research/Second_paper_material/Data2023")
species_names2 <- list.files()
z2=lapply(species_names2, read.delim) #read in all of 16 datasets
NN=700 #number of m/z values
mass=seq(50,NN,1) #m/z
MM=3240 #number of scans
ndim=length(species_names)
ndim2=length(species_names2)
#Scan number:m/z:intensity values for 134 samples
M=list()
for(i in 1:ndim){
colnames(z[[i]])="mass"
#remove commas
z[[i]]=data.frame(do.call("rbind", strsplit(as.character(z[[i]]$mass), ",",
fixed = TRUE)))
z[[i]]=data.frame(lapply(z[[i]],as.numeric))
colnames(z[[i]])=c("scan",as.character(seq(50,NN,1)))
z[[i]]=z[[i]] %>% slice(1:MM) #selects the first MM rows
M[[i]]=z[[i]]
}
#Scan number:m/z:intensity values for 16 samples
M2=list()
for(i in 1:ndim2){
colnames(z2[[i]])="mass"
#remove commas
z2[[i]]=data.frame(do.call("rbind", strsplit(as.character(z2[[i]]$mass), ",",
fixed = TRUE)))
z2[[i]]=data.frame(lapply(z2[[i]],as.numeric))
colnames(z2[[i]])=c("scan",as.character(seq(50,NN,1)))
z2[[i]]=z2[[i]] %>% slice(1:MM) #selects the first MM rows
M2[[i]]=z2[[i]]
}
M_new=c(M,M2)
N=length(M_new)
species_n2=rep(2,16)
species_names_new=c(species_names,species_names2)
y=c(species_n,species_n2)
y=factor(y,labels=c("A","B"))
#####################################################################################
#Preprocessing
#Detect the significant peaks as local max above four times the signal to noise ratio
MZ=151 #number of m/z values to use
#Create Chromatograms for each sample and each m/z value inside each sample
sample_list=list()
sample_location_list=list()
suppressWarnings({
for (i in 1:N){
S=list()
for(j in 1:MZ){
log_int=ifelse(M_new[[i]][,j+1]>0, log10(M_new[[i]][,j+1]),0)
S[[j]] = createMassSpectrum(mass=seq(1,MM,1), log_int,
metaData=list(name="Chrom"))
}
chrom = smoothIntensity(S, method="MovingAverage",halfWindowSize=5)#smooth the data
chrom = removeBaseline(chrom, method="SNIP") #remove the baseline
# Put the processed chromatograms back into a dataframe
processed_chrom_list=list()
for (k in 1:MZ){
processed_chrom_list[[k]] = as.numeric(intensity(chrom[[k]]))
}
processed_mass_dataframe = as.data.frame(do.call(rbind, processed_chrom_list))
Ma=max(processed_mass_dataframe)
Mi=min(processed_mass_dataframe)
#Normalize across sample
processed_mass_dataframe = t((processed_mass_dataframe - Mi)/(Ma - Mi))
processed_mass_dataframe = as.data.frame(processed_mass_dataframe)
S2=list()
for(t in 1:MZ){
S2[[t]] = createMassSpectrum(mass=seq(1,MM,1),
intensity=processed_mass_dataframe[, t],
metaData=list(name="Chrom_normalized"))
}
peaks = detectPeaks(S2, method="MAD", halfWindowSize=20, SNR=4)
peak_list=list()
for (tt in 1:MZ){
v=numeric(MM)
scan_number=mass(peaks[[tt]])
v[scan_number] = intensity(peaks[[tt]])
peak_list[[tt]] = v
}
processed_peaks = t(as.data.frame(do.call(rbind, peak_list)))
row.names(processed_peaks)=c(paste0("R", 1:MM))
colnames(processed_peaks)=c(paste0("M", 50:(MZ+50-1)))
processed_peaks2=as.data.frame(processed_peaks) %>%
mutate(bin = cut(seq(1,MM,1),breaks=180,dig.lab = 6)) %>% # bin the peaks by scan #
group_by(bin) %>%
summarise_all(max)
sample_list[[i]] = processed_peaks2
sample_location_list[[i]] = processed_peaks
}
})
#Scan # and mass spectrum for each sample
mass_scan_list_loc=list()
for(i in 1:N){
sm_loc=as.numeric(unlist(sample_location_list[[i]]))
mass_scan_list_loc[[i]]=sm_loc
}
#Sample vs mass/scan numbers
#Put the samples into a dataframe
data_mass_scan_loc = do.call(rbind, mass_scan_list_loc)
bin2=as.character(seq(1,3240,1))
MS=as.character(seq(50,(MZ+50-1),1))
colnames(data_mass_scan_loc) = paste(outer(bin2, MS, paste, sep = ';'))
data_mass_scan_new=as.data.frame(ifelse(data_mass_scan_loc > 0,1,0))
MZ_name=c(paste0(";",50:(MZ+50-1)))
MZ_name2=c(paste0(50:(MZ+50-1)))
MZ_name3=c(paste0(".",50:(MZ+50-1)))
bin3 = unique(cut(seq(1,MM,1),breaks=180,dig.lab = 6))
#Finding the endpoints in bin4 is based on code found in
#"Obtain endpoints from interval that is factor variable"
#https://stackoverflow.com/questions/40665240/obtain-endpoints-from-interval-that-is-factor-variable
#Stack Overflow:
bin4=unique(as.numeric(unlist( strsplit( gsub( "[][(]" , "", levels(bin3)) , ","))))
#Determine the mean scan number for each bin and each
#m/z value across the training samples
mean_scan_name = list()
for (i in 1:MZ){
data_mass_scan_new2 = data_mass_scan_new %>% select(ends_with(MZ_name[i]))
scan_name=sub(basename(colnames(data_mass_scan_new2)),pattern = MZ_name[i],
replacement = "" , fixed = TRUE)
colnames(data_mass_scan_new2) = scan_name #name the columns with the scan numbers
count_elements=function(x){
y=sum(x)
}
#Count the number of elements for each scan number across the training samples
n_elements=as.numeric(apply(as.matrix(data_mass_scan_new2),2,count_elements))
#Repeat the nonzero scan numbers with its frequency
vec=as.numeric(rep(colnames(data_mass_scan_new2), times=n_elements))
scan_dataframe=list() #scan numbers for each bin for the ith m/z value selected
indd1=1:floor(bin4[2])
ind_L1=which(vec %in% indd1)
Lt1=vec[ind_L1]
#Mean number of scan number in the first bin
scan_mean1=if(length(Lt1)>0){round(mean(Lt1))
}else {-1}
DD1=data.frame(matrix(ncol = 1, nrow = 1))
colnames(DD1)= paste(outer(as.character(scan_mean1),MZ_name2[i], paste, sep = ';'))
scan_dataframe[[1]]=DD1
for (j in 2:(length(bin4)-1)){
indd=(floor(bin4[j])+1):floor(bin4[j+1])
ind_L=which(vec %in% indd)
Lt=vec[ind_L]
#Mean number of scan number in each bin
scan_mean=if(length(Lt)>0){round(mean(Lt))
}else {-j} #to give empty columns a unique name
DD=data.frame(matrix(ncol = 1, nrow = 1))
colnames(DD)= paste(outer(as.character(scan_mean),MZ_name2[i],
paste, sep = ';'))
scan_dataframe[[j]]=DD
}
#The mean scan number for each bin of scans for the ith m/z value and each sample
mean_scan_name[[i]]=do.call(cbind, scan_dataframe)
}
# To get the column names of the interleaved m/z values and scan numbers as columns
data_mass_scan_interl=do.call(cbind, mean_scan_name)
names_new=colnames(data_mass_scan_interl)
#Scan# and mass spectrum for each sample
mass_scan_list=list()
for(i in 1:N){
sm=as.numeric(unlist(sample_list[[i]]))[181:(180*(MZ+1))]
mass_scan_list[[i]]=sm
}
#Sample vs mass/scan numbers
data_preprocessed = do.call(rbind, mass_scan_list) #put the samples into a dataframe
colnames(data_preprocessed ) = names_new
data_preprocessed = as.data.frame(data_preprocessed )
count_nonzero=function(x){(length(which(x > 0)))/(dim(data_preprocessed)[2])}
#Ratio of nonzero feature values for each observation
Perc_Nnonzero=apply(data_preprocessed ,1,count_nonzero)
data_preprocessed = data_preprocessed %>% mutate(Perc_Nnonzero)
colnames(data_preprocessed) = make.names(colnames(data_preprocessed))
#Removing variables with near zero variance, nearZeroVar, is based on the book
#"Applied Predictive Modeling" by Kuhn, M. and Johnson, K.,
#Springer, New York, 2013
#Detect features with near zero variance
near_zero_variance = nearZeroVar(data_preprocessed)
#Remove features with near zero variance
data_preprocessed = data_preprocessed[, -near_zero_variance]
dim(data_preprocessed) #new dimension
## [1] 150 8204
#Contains all samples with the selected features and corresponding y-values (A or B)
training_transformed=data.frame(data_preprocessed,y=y)
#########################################################################################
#Machine learning
#The machine learning R-code is based on the mlr3 library and the book
## https://mlr3book.mlr-org.com/, "Flexible and Robust Machine Learning Using mlr3 in R"
#by Lang, M. et al
#Create a task
task=as_task_classif(training_transformed, target = "y", positive="B")
#Using stratified random sampling for each resampling
task$col_roles$stratum = "y"
########################################################################################
#Nested resampling, elastic net, generalized testing error for
#tuning the penalty parameter alpha
set.seed(99)
inner_loop = rsmp("cv", folds = 10) # inner loop for nested CV
outer_loop = rsmp("cv", folds = 5) # outer loop for nested CV
outer_loop$instantiate(task)
#Create the learner
learner = lrn("classif.cv_glmnet", id = "glmnet", predict_type="prob")
graph = po("removeconstants")%>>% po("scale") %>>% learner
plot(graph)
graph_learner = as_learner(graph)
#graph_learner$param_set$ids()
#glmnet
graph_learner$param_set$values$glmnet.alpha = to_tune(p_dbl(0,1))
#graph_learner$param_set$values$glmnet.s = to_tune(p_dbl(0,1))
graph_learner$id = "graph_learner"
future::plan("multisession")
at = AutoTuner$new(
learner = graph_learner,
resampling = inner_loop,
measure = msr("classif.ce"),
terminator = trm("evals", n_evals = 20),
tuner = tnr("random_search"),
store_models = TRUE)
rr = resample(task = task, learner = at, resampling = outer_loop,
store_models = TRUE)
#extract_inner_tuning_results(rr)
#Unbiased prediction performance (testing error)
rr$aggregate()
## classif.ce
## 0.1133333
rr$aggregate(msr("classif.auc"))
## classif.auc
## 0.944
######################################################################
#Tune alpha in the elastic net model on the 150 training data
set.seed(99)
resampling_cv10 = rsmp("cv", folds = 10) #For 10-fold cross validation
resampling_cv10$instantiate(task)
#Create the learner
learner = lrn("classif.cv_glmnet", id = "glmnet", predict_type="prob")
graph = po("removeconstants") %>>% po("scale") %>>% learner
plot(graph)
graph_learner = as_learner(graph)
graph_learner$param_set$ids()
## [1] "removeconstants.ratio" "removeconstants.rel_tol"
## [3] "removeconstants.abs_tol" "removeconstants.na_ignore"
## [5] "removeconstants.affect_columns" "scale.center"
## [7] "scale.scale" "scale.robust"
## [9] "scale.affect_columns" "glmnet.alignment"
## [11] "glmnet.alpha" "glmnet.big"
## [13] "glmnet.devmax" "glmnet.dfmax"
## [15] "glmnet.epsnr" "glmnet.eps"
## [17] "glmnet.exclude" "glmnet.exmx"
## [19] "glmnet.fdev" "glmnet.foldid"
## [21] "glmnet.gamma" "glmnet.grouped"
## [23] "glmnet.intercept" "glmnet.keep"
## [25] "glmnet.lambda.min.ratio" "glmnet.lambda"
## [27] "glmnet.lower.limits" "glmnet.maxit"
## [29] "glmnet.mnlam" "glmnet.mxitnr"
## [31] "glmnet.mxit" "glmnet.nfolds"
## [33] "glmnet.nlambda" "glmnet.offset"
## [35] "glmnet.parallel" "glmnet.penalty.factor"
## [37] "glmnet.pmax" "glmnet.pmin"
## [39] "glmnet.prec" "glmnet.predict.gamma"
## [41] "glmnet.relax" "glmnet.s"
## [43] "glmnet.standardize" "glmnet.standardize.response"
## [45] "glmnet.thresh" "glmnet.trace.it"
## [47] "glmnet.type.gaussian" "glmnet.type.logistic"
## [49] "glmnet.type.measure" "glmnet.type.multinomial"
## [51] "glmnet.upper.limits"
#glmnet
graph_learner$param_set$values$glmnet.alpha = to_tune(p_dbl(0,1))
graph_learner$id = "graph_learner"
RS = tnr("random_search")
future::plan("multisession")
#using stratified random sampling inside each fold
instance = tune(
tuner = RS,
task = task,
learner = graph_learner,
resampling = resampling_cv10,
measure = msr("classif.ce"),
term_evals = 20
)
instance$result_y
## classif.ce
## 0.1133929
instance$result
#######################################################################
#Tune the overall penalty parameter in the elastic net model
set.seed(99)
dim_en = dim(training_transformed)[2]-1
cv_lambda=cv.glmnet(as.matrix(training_transformed[,1:dim_en]),y,
alpha=instance$result$glmnet.alpha,family="binomial")
plot(cv_lambda)
best_lambda=cv_lambda$lambda.min
elastic_mod=glmnet(as.matrix(training_transformed[,1:dim_en]),y,
alpha=instance$result$glmnet.alpha,family="binomial")
elastic_coef = predict(elastic_mod, type = "coefficients",
alpha=instance$result$glmnet.alpha,
s = best_lambda,family="binomial")
ind_p = which(c(elastic_coef[,1] > 0))
ind_n = which(c(elastic_coef[,1] < 0))
var_p=colnames(training_transformed[,ind_p - 1])
var_n = colnames(training_transformed[,ind_n - 1])
Av=c(var_p,var_n) #non-zero coefficients
AvE = data.frame(Av)