1 Introduction

This vignette illustrates how to read and input your own data to the SIAMCAT package. We will cover reading in text files from the disk, formatting them and using them to create an object of siamcat-class.

The siamcat-class is the centerpiece of the package. All of the input data and result are stored inside of it. The structure of the object is described below in the siamcat-class object section.

2 Loading your data into R

2.1 SIAMCAT input

Generally, there are three types of input for SIAMCAT:

2.1.1 Features

The features should be a matrix, a data.frame, or an otu_table, organized as follows:

features (in rows) x samples (in columns).

Sample_1 Sample_2 Sample_3 Sample_4 Sample_5
Feature_1 0.59 0.71 0.78 0.61 0.66
Feature_2 0.00 0.02 0.00 0.00 0.00
Feature_3 0.02 0.00 0.00 0.00 0.20
Feature_4 0.34 0.00 0.13 0.07 0.00
Feature_5 0.06 0.16 0.00 0.00 0.00

Please note that SIAMCAT is supposed to work with relative abundances. Other types of data (e.g. counts) will also work, but not all functions of the package will result in meaningful outputs.

An example of a typical feature file is attached to the SIAMCAT package, containing data from a publication investigating the microbiome in colorectal cancer (CRC) patients and controls (the study can be found here: Zeller et al). The metagenomics data were processed with the MOCAT pipeline, returning taxonomic profiles on the species levels (specI):

library(SIAMCAT)  <- system.file(
    package = "SIAMCAT"

One way to load such data into R could be the use of read.table

(Beware of the defaults in R! They are not always useful…)

feat <- read.table(, sep='\t',
    header=TRUE, quote='',
    stringsAsFactors = FALSE, check.names = FALSE)
# look at some features
feat[110:114, 1:2]
##                                CCIS27304052ST-3-0 CCIS15794887ST-4-0
## Bacteroides caccae [h:1096]          1.557937e-03       1.761949e-03
## Bacteroides eggerthii [h:1097]       2.734527e-05       4.146882e-05
## Bacteroides stercoris [h:1098]       1.173786e-03       2.475838e-03
## Bacteroides clarus [h:1099]          4.830533e-04       4.589747e-06
## Methanohalophilus mahii [h:11]       0.000000e+00       0.000000e+00

2.1.2 Metadata

The metadata should be either a matrix or a data.frame.

samples (in rows) x metadata (in columns):

Age Gender BMI
Sample_1 52 1 20
Sample_2 37 1 18
Sample_3 66 2 24
Sample_4 54 2 26
Sample_5 65 2 30

The rownames of the metadata should match the colnames of the feature matrix.

Again, an example of such a file is attached to the SIAMCAT package, taken from the same study:  <- system.file(
    package = "SIAMCAT"

Also here, the read.table can be used to load the data into R.

meta <- read.table(, sep='\t',
    header=TRUE, quote='',
    stringsAsFactors = FALSE, check.names = FALSE)
##                    age gender bmi diagnosis localization crc_stage fobt
## CCIS27304052ST-3-0  52      1  20         0           NA         0    0
## CCIS15794887ST-4-0  37      1  18         0           NA         0    0
## CCIS74726977ST-3-0  66      2  24         1           NA         0    0
## CCIS16561622ST-4-0  54      2  26         0           NA         0    0
## CCIS79210440ST-3-0  65      2  30         0           NA         0    1
## CCIS82507866ST-3-0  57      2  24         0           NA         0    0
##                    wif_test
## CCIS27304052ST-3-0        0
## CCIS15794887ST-4-0        0
## CCIS74726977ST-3-0       NA
## CCIS16561622ST-4-0        0
## CCIS79210440ST-3-0        0
## CCIS82507866ST-3-0        0

2.1.3 Label

Finally, the label can come in different three different flavours:

  • Named vector: A named vector containing information about cases and controls. The names of the vector should match the rownames of the metadata and the colnames of the feature data. The label can contain either the information about cases and controls either
    • as integers (e.g. 0 and 1),
    • as characters (e.g. CTR and IBD), or
    • as factors.
  • Metadata column: You can provide the name of a column in the metadata for the creation of the label. See below for an example.

  • Label file: SIAMCAT has a function called read.label, which will create a label object from a label file. The file should be organized as follows:
    • The first line is supposed to read: #BINARY:1=[label for cases];-1=[label for controls]
    • The second row should contain the sample identifiers as tab-separated list (consistent with feature and metadata).
    • The third row is then supposed to contain the actual class labels (tab-separated): 1 for each case and -1 for each control.

    An example file is attached to the package again, if you want to have a look at it.

For our example dataset, we can create the label object out of the metadata column called diagnosis:

label <- create.label(meta=meta, label="diagnosis",
    case = 1, control=0)

When we later plot the results, it might be nicer to have names for the different groups stored in the label object (instead of 1 and 0). We can also supply them to the create.label function:

label <- create.label(meta=meta, label="diagnosis",
    case = 1, control=0,
    p.lab = 'cancer', n.lab = 'healthy')
## Label used as case:
##    1
## Label used as control:
##    0
## + finished create.label.from.metadata in 0.001 s
## healthy  cancer 
##      -1       1

If you have no label information for your dataset, you can still create a SIAMCAT object from your features alone. The SIAMCAT object without label information will contain a TEST label that can be used for making holdout predictions. Other functions, e.g. model training, will not work on such an object.

2.2 LEfSe format files

LEfSe is a tool for identification of associations between micriobial features and up to two metadata. LEfSe uses LDA (linear discriminant analysis).

LEfSe input file is a .tsv file. The first few rows contain the metadata. The following row contains sample names and the rest of the rows are occupied by features. The first column holds the row names:

label healthy healthy healthy cancer cancer
age 52 37 66 54 65
gender 1 1 2 2 2
Sample_info Sample_1 Sample_2 Sample_3 Sample_4 Sample_5
Feature_1 0.59 0.71 0.78 0.61 0.66
Feature_2 0.00 0.02 0.00 0.00 0.00
Feature_3 0.02 0.00 0.00 0.00 0.00
Feature_4 0.34 0.00 0.43 0.00 0.00
Feature_5 0.56 0.56 0.00 0.00 0.00

An example of such a file is attached to the SIAMCAT package:<- system.file(
    package = "SIAMCAT"

SIAMCAT has a dedicated function to read LEfSe format files. The read.lefse function will read in the input file and extract metadata and features:

meta.and.features <- read.lefse(,
    rows.meta = 1:6, row.samples = 7)
meta <- meta.and.features$meta
feat <- meta.and.features$feat

We can then create a label object from one of the columns of the meta object and create a siamcat object:

label <- create.label(meta=meta, label="label", case = "cancer")
## Label used as case:
##    cancer
## Label used as control:
##    healthy
## + finished create.label.from.metadata in 0.005 s

2.3 metagenomeSeq format files

metagenomeSeq is an R package to determine differentially abundant features between multiple samples.

There are two ways to input data into metagenomeSeq:

  1. two files, one for metadata and one for features - those can be used in SIAMCAT just like described in SIAMCAT input with read.table:  <- system.file(
    package = "metagenomeSeq"
feat <- read.table(, sep='\t',
    header=TRUE, quote='', row.names = 1,
    stringsAsFactors = FALSE, check.names = FALSE
  1. BIOM format file, that can be used in SIAMCAT as described in the following section

2.4 BIOM format files

The BIOM format files can be added to SIAMCAT via phyloseq. First the file should be imported using the phyloseq function import_biom. Then a phyloseq object can be imported as a siamcat object as descibed in the next section.

2.5 Creating a siamcat object of a phyloseq object

The siamcat object extends on the phyloseq object. Therefore, creating a siamcat object from a phyloseq object is really straightforward. This can be done with the siamcat constructor function. First, however, we need to create a label object:

data("GlobalPatterns") ## phyloseq example data
label <- create.label(meta=sample_data(GlobalPatterns),
    label = "SampleType",
    case = c("Freshwater", "Freshwater (creek)", "Ocean"))
## Label used as case:
##    Freshwater,Freshwater (creek),Ocean
## Label used as control:
##    rest
## + finished create.label.from.metadata in 0.003 s
# run the constructor function
siamcat <- siamcat(phyloseq=GlobalPatterns, label=label)
## + starting
## +++ checking overlap between labels and features
## + Keeping labels of 26 sample(s).
## +++ checking sample number per class
## Data set has only 8 training examples of class Case .
## Note that a dataset this small/skewed is not necessarily suitable for analysis in this pipeline.
## +++ checking overlap between samples and metadata
## + finished in 0.659 s

3 Creating a siamcat-class object

The siamcat-class is the centerpiece of the package. All of the is stored inside of the object: internal make-up of a siamcat object

In the figure above, rectangles depict slots of the object and the class of the object stored in the slot is given in the ovals. There are two obligatory slots -phyloseq (containing the metadata as sample_data and the original features as otu_table) and label - marked with thick borders.

The siamcat object is constructed using the siamcat() function. There are two ways to initialize it:

Please note that you have to provide either feat or phyloseq and that you cannot provide both.

In order to explain the siamcat object better we will show how each of the slots is filled.

3.1 phyloseq, label and orig_feat slots

The phyloseq and label slots are obligatory.

  • The phyloseq slot is an object of class phyloseq, which is described in the help file of the phyloseq class. Help can be accessed by typing into R console: help('phyloseq-class').
    • The otu_table slot in phyloseq -see help('otu_table-class')- stores the original feature table. For SIAMCAT, this slot can be accessed by orig_feat.
  • The label slot contains a list. This list has a specific set of entries -see help('label-class')- that are automatically generated by the read.label or create.label functions.

The phyloseq, label and orig_feat are filled when the siamcat object is first created with the constructor function. construction

3.2 All the other slots

Other slots are filled during the run of the SIAMCAT workflow: