NetMoss2 is the new version of NetMoss, which is a tool developed for integrating large-scale data and identifying disease associated biomarkers based on network algorithm.
In this new version, both single file and multiple files are supported as input. Also, we provide visualization to clearly illustrate correponding results.
For more information, please see paper "Large-scale microbiome data integration enables robust biomarker identification" published in Nature Computational Science.
Install NetMoss2 with remotes
install.packages("remotes")
remotes::install_github("xiaolw95/NetMoss2")
library(NetMoss2)
For some R packages like rsparcc, installation from CRAN may be complicated. We recommend install them from github directly.
For example:
install_github("MPBA/r-sparcc")
The NetMoss function is used to calculate NetMoss score of significant bacteria between case and control groups. Users are demanded to provide four directories or files as follows:
nodes_result = NetMoss(case_dir = case_dir,
control_dir = control_dir,
net_case_dir = net_case_dir,
net_control_dir = net_control_dir)
case_dir: the directory or a single file of case data.
control_dir: the directory or a single file of control data.
net_case_dir: the directory or a single file of case network.
net_control_dir: the directory or a single data of control network.
Other arguments related to model division can be referred to original WGCNA publications and papers:
WGCNA: an R package for weighted correlation network analysis
Defining clusters from a hierarchical cluster tree: the Dynamic Tree Cut package for R
Eigengene networks for studying the relationships between co-expression modules
Abundance or network matrix should be included in the input.
case_dir or control_dir includes abundance matrix which refers to the relative abundance of case or contol samples, with the row as bacteria and the column as samples. Abundance file can be processed from raw sequence using QIIME2, MetaPhlAn3 or other tools.
NOTE: be sure that set header = T when you read the data and leave the name of the first column as blank as the following example:
| sample1 | sample2 | sample3 | |
|---|---|---|---|
| taxon1 | 60 | 20 | 10 |
| taxon2 | 30 | 77 | 89 |
| taxon3 | 0 | 23 | 15 |
| ... ... |
net_case_dir or net_control_dir includes network matrix which refers to the adjacency matrix of correltaion between the bacteria. Microbial correlation can be deduced from any tools for which SparCC or SPIEC-EASI are especially recommended.
| taxon1 | taxon2 | taxon3 | |
|---|---|---|---|
| taxon1 | 1 | -0.3 | 0.5 |
| taxon2 | -0.3 | 1 | 0.67 |
| taxon3 | 0.5 | 0.67 | 1 |
| ... ... |
For convenience, we also provide a netBuild function to build microbial networks from abundance tables. To use this function, users are asked to provide abundance directories (contain case and control abundance tables). Network matrix will be output to the same parent directories automatically. For single file usage, users are asked to provided the abundance matrix only.
NOTE: The netBuild function will creat "net_case_dir" and "net_control_dir" directories and output the network results into them. If the same directories exist, files will be overwritten.
First, make sure that you have the package rsparcc installed.
install_github("MPBA/r-sparcc")
Next, you can load rsparcc and build the network.
library(rsparcc)
netBuild(case_dir = case_dir,
control_dir = control_dir,
method = "sparcc")
case_dir: the directory or a single file of case data.
control_dir: the directory or a single file of control data.
method: the method to build networks. "sparcc" and "pearson" strategy are provided to choose.
The output of the NetMoss function is a list contains NetMoss score and integrated results, which could be used in the downstream analysis.
The first list: nodes_result[[1]] is the NetMoss score and p value of each taxon;
The second and the third list: nodes_result[[2]] and nodes_result[[3]] are integrated networks constructed from case and control data;
The 4th and thd 5th list: nodes_result[[4]] and nodes_result[[5]] are case and control data from the input.
The first list contains a table of NetMoss score for each taxon:
| taxon_names | NetMoss_score | p.val | p.adj |
|---|---|---|---|
| taxon1 | 0.98 | 5.703785e-09 | 2.335836e-08 |
| taxon2 | 0.7 | 1.467413e-04 | 2.629116e-04 |
| taxon3 | 0.32 | 2.018237e-04 | 3.542211e-04 |
| ... ... |
taxon_names: the name of the bacteria.
NetMoss_score: the NetMoss of the bacteria gets.
p.val: the P value for the NetMoss score.
p.adj: the adjust P value for the NetMoss score.
In this part, we provide a function to illustrate the results. The netPlot function will output two kind of images. The one is the visualization of the important NetMoss score, the other is the paired networks to demostrated the difference of structure between case and control networks.
netPlot(result = nodes_result,
num.top = 5,
num.score = 30,
e.th = 0.4,
my.layout = layout_components,
my.label = TRUE)
result: the result from NetMoss function. This is a list contained NetMoss score and integrated networks.
num.top: a numerical parameter. the number of top taxon to be highlighted in the paired networks.
num.score: a numerical parameter. the number of taxon to be plotted in the NetMoss score barplot and point plot.
e.th: the threshold of microbial correlations in the networks. Edges greater than this threshold are shown in the networks.
my.layout: the layout of network. See package "igraph" for more information.
my.label a logical parameter. If TRUE then the label of the nodes will be plotted.
The image of NetMoss score demostrates the NetMoss score and average abundance of top 30 bacteria, which can be modified using the parameter num.score:
Also, there is a image to demostrate the the difference of structure between case and control networks.
The image highlights the top 5 taxon with the highest NetMoss score. The taxon number can be modified by hand using the parameter num.top:
In this section, we provide a pipeline to classify case and control groups based on the NetMoss markers. Iterative training and 10-fold cross validation stpes are implemented to guarantee the markers contain network and abundance informations. For this reason, it will take a long time to process the real datasets which contain large samples. Please be patient.
netROC(case_dir = case_dir,
control_dir = control_dir,
marker = marker,
metadata = metadata,
plot.roc = TRUE,
train.num = 20)
case_dir: the directory or a single file of case data.
control_dir: the directory or a single file of control data.
marker: a table of combined markers identified by NetMoss.
metadata: a table of clinical informations for all studies.
plot.roc: a logical parameter. If TRUE then the combined ROC of the result of classification will be plotted.
train.num: a numerical parameter which refers to trainning times of the model. By default, it is set to 20.
First of all, efficient markers should be selected manually from the NetMoss result by users. Generally, we recommend a less strict threshold for the sparse network. Also, a metadata file contains disease or health information for each sample needs to be inculded. The format should be like this:
| sample_id | type | study |
|---|---|---|
| SRRXXXXX | disease | study1 |
| SRRXXXXX | disease | study2 |
| SRRXXXXX | healthy | study1 |
| ... ... |
After preparing the two files, classification can be realized using the function netROC
The result of the classfication contains a table includes true positive rate and false positive rate:
| threhold | TPR | FPR |
|---|---|---|
| 0 | 1 | 1 |
| 0.01 | 0.97 | 0.99 |
| 0.03 | 0.9 | 0.87 |
| ... ... |
Also, it contains a table of markers used in the classification:
| Name | NetMoss_score |
|---|---|
| tax1 | 1 |
| tax2 | 0.9 |
| tax3 | 0.84 |
| ... ... |
A combined ROC will be ploted if the parameter plot.roc is set to be true.
We have provided a small dataset to test the R package. In our testData directory, both abundance files and network files, as well as the metadata, are included.
- Download from the testData directory (https://github.com/xiaolw95/NetMoss2/tree/main/testData) directly.
Or get the dataset usinggit clonecommond inLinux:
git clone https://github.com/xiaolw95/NetMoss2.git
cd NetMoss2/testData
- After getting the dataset, the NetMoss score can be easily calculated using the
NetMossfunction:
#load package
library(NetMoss2)
#setwd('path-to-testData-directory') ####set the directory to testthat
#read directory
case_dir = paste0(getwd(),"/case_dir")
control_dir = paste0(getwd(),"/control_dir")
net_case_dir = paste0(getwd(),"/net_case_dir")
net_control_dir = paste0(getwd(),"/net_control_dir")
#construct networks ####if files exist, skip
#library(rsparcc)
#netBuild(case_dir = case_dir,
# control_dir = control_dir,
# method = "sparcc")
#calculate NetMoss score
nodes_result = NetMoss(case_dir = case_dir,
control_dir = control_dir,
net_case_dir = net_case_dir,
net_control_dir = net_control_dir)
result = nodes_result[[1]] ####NetMoss score result
#plot networks
netPlot(nodes_result) ####image saved
#plot roc
#trim markers
marker = data.frame(result[which(result$p.adj < 0.05),])
marker = data.frame(marker[which(marker$NetMoss_Score > 0.3),]) ####marker selection
rownames(marker) = marker$taxon_names
#construct metadata ######if file exists, skip
#metadata
#case = nodes_result[[4]]
#control = nodes_result[[5]]
#metadata = data.frame(sample_id = c(colnames(case[,-1]),
# colnames(control[,-1])),
# type = c(rep("disease",length(colnames(case[,-1]))),
# rep("healthy",length(colnames(control[,-1])))))
#metadata$sample_id = as.character(metadata$sample_id)
#metadata$type = as.factor(metadata$type)
#rownames(metadata) = metadata$sample_id
metadata = read.table("metadata.txt",header = T,sep = '\t',row.names = 1)
myROC = netROC(case_dir = case_dir,
control_dir = control_dir,
marker = marker,
metadata = metadata,
plot.roc = TRUE,
train.num = 20) ####image saved
If users only have single file for case and control groups, NetMoss2 can also be used to identify significant biomarkes.
#load package
library(NetMoss2)
#setwd("your-directory")
#load dataset
data(testData)
#contruct networks ####if files exist, skip
#library(rsparcc)
#netBuild(case_dir = mydata[[1]],
# control_dir = mydata[[2]],
# method = "sparcc")
#calculate NetMoss score
nodes_result = NetMoss(case_dir = mydata[[1]],
control_dir = mydata[[2]],
net_case_dir = mydata[[3]],
net_control_dir = mydata[[4]])
result = nodes_result[[1]] ####NetMoss score result
#plot networks
netPlot(nodes_result) ####image saved
#plot roc
#trim markers
marker = data.frame(result[which(result$p.adj < 0.05),])
marker = data.frame(marker[which(marker$NetMoss_Score > 0.3),]) ####marker selection
rownames(marker) = marker$taxon_names
#construct metadata ######if file exists, skip
#metadata
#case = nodes_result[[4]]
#control = nodes_result[[5]]
#metadata = data.frame(sample_id = c(colnames(case[,-1]),
# colnames(control[,-1])),
# type = c(rep("disease",length(colnames(case[,-1]))),
# rep("healthy",length(colnames(control[,-1])))))
#metadata$sample_id = as.character(metadata$sample_id)
#metadata$type = as.factor(metadata$type)
#rownames(metadata) = metadata$sample_id
metadata = mydata[[5]]
myROC = netROC(case_dir = mydata[[1]],
control_dir = mydata[[2]],
marker = marker,
metadata = metadata,
plot.roc = TRUE,
train.num = 20) ####image saved