In pooranis/maaslin: Maaslin

MaAsLin User Guide v3.1

February 2015

Ayshwarya Subramanian, Timothy Tickle, Levi Waldron, George Weingart, Lauren McIver and Curtis Huttenhower

Introduction

MaAsLin is a multivariate statistical framework that finds associations between clinical metadata and potentially high-dimensional experimental data. MaAsLin performs boosted additive general linear models between one group of data (metadata/the predictors) and another group (in our case relative taxonomic abundances/the response). In our context we use it to discover associations between clinical metadata and microbial community relative abundance or function; however, it is applicable to other data types.

Metagenomic data are sparse, and boosting is used to select metadata that show some potential to be useful in a linear model between the metadata and abundances. In the context of metadata and community abundance, a sample's metadata is boosted for one Operational Taxonomic Unit (OTU) (Yi). The metadata that are selected by boosting are then used in a general linear model, with each combination of metadata (as predictors) and OTU abundance (as response variables). This occurs for every OTU and metadata combination. Given we work with proportional data, the Yi (abundances) are arcsin(sqrt(Yi)) transformed. A final formula is as follows:

For more information about maaslin please visit http://huttenhower.sph.harvard.edu/maaslin.

Installation

Download MaAsLin

MaAsLin can be downloaded in two ways:

• Download a compressed set of files.

• Create a clone of the repository on your computer with the command:

  hg clone https://bitbucket.org/biobakery/maaslin

Note: Creating a clone of the repository requires Mercurial to be installed. Once the repository has been cloned upgrading to the latest release of MaAsLin is simple. Just type hg pull -u from within the repository which will download the latest release.

Required libraries

R Libraries: Several libraries need to be installed in R these are the following:

• agricolae, gam, gamlss, gbm, glmnet, inlinedocs, logging, MASS, nlme, optparse, outliers, penalized, pscl, robustbase, testhat, vegan

You can install them by typing R in a terminal and using the install.packages command:

      install.packages(c('agricolae', 'gam', 'gamlss', 'gbm', 'glmnet', 'inlinedocs', 'logging', 'MASS', 'nlme', 'optparse', 'outliers', 'penalized', 'pscl', 'robustbase', 'testthat','vegan'))

How to Run

Input Files

There are 3 input files for each project, the "*read.config" file, the "*.pcl" file, and the "*.R" script. Details of each file follow:

1. PCL File

Required input file *.pcl. A PCL file is the file that contains all the data and metadata. This file is formatted so that metadata/data (otus or bugs) are rows and samples are columns. All metadata rows should come first before any abundance data. The file should be a tab delimited text file with the extension ".pcl".

2. Read Config File

Required input file *.read.config. A read config file allows one to indicate what data is read from a PCL file without having to change the pcl file or change code. This means one can have a pcl file which is a superset of metadata and abundances which includes data you are not interested in for the run. This file is a text file with ".read.config" as an extension. This file is later described in detail in section Process Flow subsection 4. Create your read.config file.

3. R Script File (Optional)

Optional input file *.R. The R script file is using a call back programming pattern that allows one to add/modify specific code to customize analysis without touching the main MaAsLin engine. A generic R script is provided "maaslin_demo2.R" and can be renamed and used for any study. The R script can be modified to add quality control or formatting of data, add ecological measurements, or other changes to the underlying data before MaAsLin runs on it. This file is not required to run MaAsLin.

Process Flow

1. Obtain your abundance or relative function table.

Abundance tables are normally derived from sequence data using Mothur, Qiime, HUMAnN, or MetaPhlAn. Please refer to their documentation for further details.

2. Obtain your metadata.

Metadata would be information about the samples in the study. For instance, one may analyze a case / control study. In this study, you may have a disease and healthy group (disease state), the sex of the patents (patient demographics), medication use (chemical treatment), smoking (patient lifestyle) or other types of data. All aforementioned data would be study metadata. This section can have any type of data (factor, ordered factor, continuous, integer, or logical variables). If a particular data is missing for a sample for a metadata please write NA. It is preferable to write NA so that, when looking at the data, it is understood the metadata is missing and it's absence is intentional and not a mistake. Often investigators are interested in genetic measurements that may also be placed in the metadata section to associate to bugs.

Please note that no special characters are allowed in the metadata header names. These names should only contain alphanumeric characters with the addition of a period and an underscore: [a-zA-Z0-9.-].

3. Format and combine your abundance table and metadata as a pcl

Please note two tools have been developed to help you! If you are working from a Qiime OTU output and have a metadata text file try using QiimeToMaaslin found at bitbucket. If you have a tab delimited file which matches the below .pcl description (for instance MetaPhlAn output) use the merge_metadata.py script provided in this project (maaslin/exec/merge_metadata.py) and documented in maaslin/doc/Merge_Metadata_Read_Me.txt.

PCL format description

i. Row 1 is expected to be sample IDs beginning the first column with a feature name to identify the row, for example "ID".

ii. Rows of metadata. Each row is one metadata, the first column entry being the name of the metadata and each following column being the metadata value for that each sample.

iii. Row of taxa/otu abundance. Each row is one taxa/otu, the first column entry being the name of the taxa/otu followed by abundances of the taxa/otu per sample.

iv. Abundances should be normalized by dividing each abundance measurement by the sum of the column (sample) abundances.

v. Here is an example of the contents of an extremely small pcl file; another example can be found in this project at maaslin/inst/extdata/maaslin_demo.pcl.

ID  Sample1 Sample2 Sample3 Sample4
metadata1   True    True    False   False
metadata2   1.23    2.34    3.22    3.44
metadata3   Male    Female  Male    Female
taxa1       0.022   0.014   0.333   0.125
taxa2       0.406   0.029   0.166   0.300
taxa3       0.571   0.955   0.500   0.575

4. Create your read.config file.

A .read.config file is a structured text file used to indicate which data in a .pcl file should be read into MaAsLin and used for analysis. This allows one to keep their *.pcl file intact while varying analysis. Hopefully, this avoids errors that may be introduced while manipulating the pcl files.

Here is an example of the contents of a *.read.config file.

Matrix: Metadata
Read_PCL_Columns: Sample2-Sample15
Read_PCL_Rows: Age-Height,Weight,Sex,Cohort-Profession

Matrix: Abundance
Read_PCL_Columns: Sample2-Sample15
Read_PCL_Rows: Bacteria-Bug100

The minimal requirement for a MaAsLin .read.config file is as follows. The Matrix: should be specified. Metadata needs to be named "Metadata" for the metadata section and "Abundance" for the abundance section. “Read_PCL_Rows:” is used to indicate which rows are data or metadata to be analyzed. Rows can be identified by their metadata/data id. Separate ids by commas. If there is a consecutive group of metadata/data a range of rows can be defined by indicating the first and last id separated by a “-“. If the beginning or ending id is missing surrounding an “–“, the rows are read from the beginning or to the end of the pcl file, respectively.

A minimal example is shown here:

Matrix: Metadata
Read\_PCL\_Rows: -Weight

Matrix: Abundance
Read_PCL_Rows: Bacteria-

With this minimal example, the delimiter of the file is assumed to be a tab, all columns are read (since they are not indicated here). Metadata are read as all rows from the beginning of the pcl file (skipping the first Sample ID row) to Weight; all data are read as all rows from Bacteria to the end of the pcl file. This example refers to the default input files given in the MaAsLin download as maaslin/inst/extdata/maaslin_demo2.*.

5. Create your R script (Optional).

The R script is used to add code that manipulates your data before analysis, and for manipulating the multifactoral analysis figure. A default “*.R” script is available with the default MaAsLin project at maaslin/inst/extdata/maaslin_demo2.R. This is an expert option and should only be used by someone very comfortable with the R language.

6. Specify Output Directory

If you have a specific directory where the results must save, then specify it's location. If not, specify the name of the directory to create with it's path. E.g. If you want results to go the folder "Results" in your current path, specify "Results". If it must go to the directory "Results" in another location, specify "Path of other location/Results".

7. Run.

On the commandline call the Maaslin.R script. Please refer to the help (-h, --help) for command line options. If running from commandline, the PCL file will need to be transposed. A script is included in Maaslin for your convenience (exec/transpose.py).

For an input file of name "input.pcl" with a read config file of name "input.read.config" and output directory name "outdir", the following is the generic command, to be executed inside the maaslin directory:

Step 1: Transposing the pcl file to a tab separated file
$ ./exec/transpose.py < input.pcl > input.tsv

Step 2: Running Maaslin
$ ./R/Maaslin.R -i input.read.config input.tsv outputdir

The following demo example is included with the package. An example call from the Maaslin folder for the demo data could be as follows.

$ ./exec/transpose.py < inst/extdata/maaslin_demo2.pcl > maaslin_demo2.tsv
$ ./R/Maaslin.R -i inst/extdata/maaslin_demo2.read.config maaslin_demo2.tsv outputdir

8. Discover amazing associations in your results!

Output Files

The following files will be generated per MaAsLin run. In the following listing the term $PROJECTNAME refers to what you named your "*.pcl" file without the extension. There are four main types of output files. Additional files are created in a folder named "QC".

1. $PROJECTNAME_log.txt

This file contains the detail for the statistical engine. This can be useful for detailed troubleshooting.

2. $PROJECTNAME-$METADATA.txt

Each metadata will have a file of associations. $METADATA is the metadata column name in the "*.pcl" file. Any associations indicated to be performed after initial variable selection (boosting) is recorded here. Included are the information from the final general linear model (performed after the boosting) and the FDR corrected p-value (q-value). Can be opened as a text file or spreadsheet.

3. $PROJECTNAME-$METADATA.pdf

Any association that had a q-value less than or equal to the given significance threshold will be plotted here (default is 0.25; can be changed using the commandline argument -d). If this file does not exist, the $PROJECTNAME-$METADATA.txt should not have an entry that is less than or equal to the threshold. Factor data is plotted as knotched box plots; continuous data is plotted as a scatter plot with a line of best fit.

4. $PROJECTNAME.txt

A collection of all entries in the $PROJECTNAME-$METADATA.pdf. Can be opened as a text file or spreadsheet.

5. QC/data.tsv

The data matrix that was read in (transposed). Useful for making sure the correct data was read in.

6. QC/data.read.config

Can be used to read in the data.tsv.

7. QC/metadata.tsv

The metadata that was read in (transposed). Useful for making sure the correct metadata was read in.

8. QC/metadata.read.config

Can be used to read in the data.tsv.

9. QC/read-Merged.tsv

The data and metadata merged (transposed). Useful for making sure the merging occurred correctly.

10. QC/read-Merged.read.config

Can be used to read in the read_merged.tsv.

11. QC/read_cleaned.tsv

The data read in, merged, and then cleaned. After this process the data is written to this file for reference if needed.

12. QC/read_cleaned.read.config

Can be used to read in read_cleaned.tsv.

13. QC/ProcessQC.txt

Contains quality control for the MaAsLin analysis. This includes information on the magnitude of outlier removal.

14. QC/Run_Parameters.txt

Contains an account of all the options used when running MaAsLin so the exact methodology can be recreated if needed.

Commandline Options

Although we recommend the use of default options, commandline arguments exist to modify both MaAsLin methodology and figures. To see an up-to-date listing of argument usage, in a terminal in the maaslin directory type ./R/Maaslin.R -h.

Example args:

-v DEBUG -d 0.1 -b 5

In this example MaAsLin is modified to produce verbose output for debugging (-v DEBUG), to change the threshold for making pdfs to a q-value equal to or less than 0.1 (-d 0.1), and to plot 5 data (bug) features in the biplot (-b 5).

All verses All

The all verses all analysis flow is a way of manipulating how metadata are used. In this method there is a group of metadata that are always evaluated, as well there are a group that are added to this one at a time. To give a more concrete example: You may have metadata cage, diet, and treatment. You may always want to have the association of abundance evaluated controlling for cage but otherwise looking at the metadata one at a time. In this way the cage metadata is the \D2forced\D3 part of the evaluation while the others are not forced and evaluated in serial. The appropriate commandline to indicate this follows:

-a -F cage

-a indicates all verses all is being used, -F indicates which metadata are forced (multiple metadata can be given comma delimited as shown here -F metadata1,metadata2,metadata3). This does not bypass the feature selection method so the metadata that are not forced are subject to feature selection and may be removed before coming to the evaluation. If you want all the metadata that are not forced to be evaluated in serial you will need to turn off feature selection and will have a final combined commandline as seen here:

-a -F cage -s none

pooranis/maaslin documentation built on May 25, 2019, 11:23 a.m.