|
0
|
1 MaAsLin User Guide v3.1
|
|
|
2 =======================
|
|
|
3
|
|
|
4 September 2013 - Updated April 2014 for Galaxy
|
|
|
5
|
|
|
6 Timothy Tickle and Curtis Huttenhower
|
|
|
7
|
|
|
8 Table of Contents
|
|
|
9 -----------------
|
|
|
10
|
|
|
11 A. Introduction to MaAsLin
|
|
|
12 B. Related Projects and Scripts
|
|
|
13 C. Installing MaAsLin
|
|
|
14 D. MaAsLin Inputs
|
|
|
15 E. Process Flow Overview
|
|
|
16 D. Process Flow Detail
|
|
|
17 G. Expected Output Files
|
|
|
18 H. Troubleshooting
|
|
|
19 I. Installation as an Automated Pipeline
|
|
|
20 J. Commandline Options (Modifying Process and Figures)
|
|
|
21
|
|
|
22 # A. Introduction to MaAsLin
|
|
|
23
|
|
|
24 MaAsLin is a multivariate statistical framework that finds
|
|
|
25 associations between clinical metadata and potentially
|
|
|
26 high-dimensional experimental data. MaAsLin performs boosted additive
|
|
|
27 general linear models between one group of data (metadata/the
|
|
|
28 predictors) and another group (in our case relative taxonomic
|
|
|
29 abundances/the response). In our context we use it to discover
|
|
|
30 associations between clinical metadata and microbial community
|
|
|
31 relative abundance or function; however, it is applicable to other
|
|
|
32 data types.
|
|
|
33
|
|
|
34 Metagenomic data are sparse, and boosting is used to select metadata
|
|
|
35 that show some potential to be useful in a linear model between the
|
|
|
36 metadata and abundances. In the context of metadata and community
|
|
|
37 abundance, a sample's metadata is boosted for one Operational
|
|
|
38 Taxonomic Unit (OTU) (Yi). The metadata that are selected by boosting
|
|
|
39 are then used in a general linear model, with each combination of
|
|
|
40 metadata (as predictors) and OTU abundance (as response
|
|
|
41 variables). This occurs for every OTU and metadata combination. Given
|
|
|
42 we work with proportional data, the Yi (abundances) are
|
|
|
43 `arcsin(sqrt(Yi))` transformed. A final formula is as follows:
|
|
|
44
|
|
|
45 
|
|
|
46
|
|
|
47 For more information about maaslin please visit
|
|
|
48 [http://huttenhower.sph.harvard.edu/maaslin](http://huttenhower.sph.harvard.edu/maaslin).
|
|
|
49
|
|
|
50
|
|
|
51 # B. Related Projects and Scripts
|
|
|
52
|
|
|
53 Other projects exist at www.bitbucket.com that may help in your
|
|
|
54 analysis:
|
|
|
55
|
|
|
56 * **QiimeToMaAsLin** is a project that reformats abundance files from
|
|
|
57 Qiime for MaAsLin. Several formats of Qiime consensus lineages are
|
|
|
58 supported for this project. To download please visit
|
|
|
59 [https://bitbucket.org/timothyltickle/qiimetomaaslin](https://bitbucket.org/timothyltickle/qiimetomaaslin).
|
|
|
60
|
|
|
61 * **merge_metadata.py** is a script included in the MaAsLin project to
|
|
|
62 generically merge a metadata file with a table of microbial (or
|
|
|
63 other) measurements. This script is located in `maaslin/src` and
|
|
|
64 is documented in `maaslin/doc/ Merge_Metadata_Read_Me.txt`.
|
|
|
65
|
|
|
66
|
|
|
67 # C. Installing MaAsLin
|
|
|
68
|
|
|
69 R Libraries: Several libraries need to be installed in R these are
|
|
|
70 the following:
|
|
|
71
|
|
|
72 * agricolae, gam, gamlss, gbm, glmnet, inlinedocs, logging, MASS, nlme, optparse, outliers, penalized, pscl, robustbase, testhat, vegan
|
|
|
73
|
|
|
74 You can install them by typing R in a terminal and using the
|
|
|
75 install.packages command:
|
|
|
76
|
|
|
77 install.packages(c('agricolae', 'gam', 'gamlss', 'gbm', 'glmnet', 'inlinedocs', 'logging', 'MASS', 'nlme', 'optparse', 'outliers', 'penalized', 'pscl', 'robustbase', 'testthat'))
|
|
|
78
|
|
|
79 # D. MaAsLin Inputs
|
|
|
80
|
|
|
81 There are 3 input files for each project, the "\*.read.config" file, the "\*.pcl" file, and the "\*.R" script. (If using the sfle automated pipeline, the "\*" in the file names can be anything but need to be identical for all three files. All three files need to be in the `../sfle/input/maasalin/input` folder only if using sfle). Details of each file follow:
|
|
|
82
|
|
|
83 ### 1\. "\*.pcl"
|
|
|
84
|
|
|
85 Required input file. A PCL file is the file that contains all the data
|
|
|
86 and metadata. This file is formatted so that metadata/data (otus or
|
|
|
87 bugs) are rows and samples are columns. All metadata rows should come
|
|
|
88 first before any abundance data. The file should be a tab delimited
|
|
|
89 text file with the extension ".pcl".
|
|
|
90
|
|
|
91 ### 2\. "\*.read.config"
|
|
|
92
|
|
|
93 Required input file. 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 **F. Process Flow Overview** subsection **4. Create your read.config file**.
|
|
|
94
|
|
|
95 ### 3\. "\*.R"
|
|
|
96
|
|
|
97 Optional input file. 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.
|
|
|
98
|
|
|
99 # E. Process Flow Overview
|
|
|
100
|
|
|
101 1. Obtain your abundance or relative function table.
|
|
|
102 2. Obtain your metadata.
|
|
|
103 3. Format and combine your abundance table and metadata as a pcl file for MaAsLin.
|
|
|
104 4. Create your read.config file.
|
|
|
105 5. Create your R script or use the default.
|
|
|
106 6. Place .pcl, .read.config, .R files in `../sfle/input/maaslin/input/` (sfle only)
|
|
|
107 7. Run
|
|
|
108 8. Discover amazing associations in your results!
|
|
|
109
|
|
|
110 # F. Process Flow Detail
|
|
|
111
|
|
|
112 ### 1\. Obtain your abundance or relative function table.
|
|
|
113
|
|
|
114 Abundance tables are normally derived from sequence data using
|
|
|
115 *Mothur*, *Qiime*, *HUMAnN*, or *MetaPhlAn*. Please refer to their documentation
|
|
|
116 for further details.
|
|
|
117
|
|
|
118 ### 2\. Obtain your metadata.
|
|
|
119
|
|
|
120 Metadata would be information about the samples in the study. For
|
|
|
121 instance, one may analyze a case / control study. In this study, you
|
|
|
122 may have a disease and healthy group (disease state), the sex of the
|
|
|
123 patents (patient demographics), medication use (chemical treatment),
|
|
|
124 smoking (patient lifestyle) or other types of data. All aforementioned
|
|
|
125 data would be study metadata. This section can have any type of data
|
|
|
126 (factor, ordered factor, continuous, integer, or logical
|
|
|
127 variables). If a particular data is missing for a sample for a
|
|
|
128 metadata please write NA. It is preferable to write NA so that, when
|
|
|
129 looking at the data, it is understood the metadata is missing and it's
|
|
|
130 absence is intentional and not a mistake. Often investigators are
|
|
|
131 interested in genetic measurements that may also be placed in the
|
|
|
132 metadata section to associate to bugs.
|
|
|
133
|
|
|
134 If you are not wanting to manually add metadata to your abundance
|
|
|
135 table, you may be interested in associated tools or scripts to help
|
|
|
136 combine your abundance table and metadata to create your pcl
|
|
|
137 file. Both require a specific format for your metadata file. Please
|
|
|
138 see the documentation for *QiimeToMaaslin* or *merge_metadata.py* (for
|
|
|
139 more details see section B).
|
|
|
140
|
|
|
141 ### 3\. Format and combine your abundance table and metadata as a pcl
|
|
|
142 file for *MaAsLin*.
|
|
|
143
|
|
|
144 Please note two tools have been developed to help you! If you are
|
|
|
145 working from a Qiime OTU output and have a metadata text file try using
|
|
|
146 *QiimeToMaaslin* found at bitbucket. If you have a tab delimited file
|
|
|
147 which matches the below .pcl description (for instance MetaPhlAn
|
|
|
148 output) use the merge_metadata.py script provided in this project
|
|
|
149 (`maaslin/src/merge_metadata.py`) and documented in
|
|
|
150 `maaslin/doc/Merge_Metadata_Read_Me.txt`.
|
|
|
151
|
|
|
152 ###PCL format description:
|
|
|
153
|
|
|
154 i. Row 1 is expected to be sample IDs beginning the first column with a feature name to identify the row, for example "ID".
|
|
|
155
|
|
|
156 ii. Rows of metadata. Each row is one metadata, the first column entry
|
|
|
157 being the name of the metadata and each following column being the
|
|
|
158 metadata value for that each sample.
|
|
|
159
|
|
|
160 iii. Row of taxa/otu abundance. Each row is one taxa/otu, the first
|
|
|
161 column entry being the name of the taxa/otu followed by abundances of
|
|
|
162 the taxa/otu per sample.
|
|
|
163
|
|
|
164 iv. Abundances should be normalized by dividing each abundance measurement by the sum of the column (sample) abundances.
|
|
|
165
|
|
|
166 v. Here is an example of the contents of an extremely small pcl file;
|
|
|
167 another example can be found in this project at
|
|
|
168 `maaslin/input/maaslin_demo.pcl`.
|
|
|
169
|
|
|
170
|
|
|
171 ID Sample1 Sample2 Sample3 Sample4
|
|
|
172 metadata1 True True False False
|
|
|
173 metadata2 1.23 2.34 3.22 3.44
|
|
|
174 metadata3 Male Female Male Female
|
|
|
175 taxa1 0.022 0.014 0.333 0.125
|
|
|
176 taxa2 0.406 0.029 0.166 0.300
|
|
|
177 taxa3 0.571 0.955 0.500 0.575
|
|
|
178
|
|
|
179
|
|
|
180 ### 4\. Create your read.config file.
|
|
|
181
|
|
|
182 A *.read.config file is a structured text file used to indicate which
|
|
|
183 data in a *.pcl file should be read into MaAsLin and used for
|
|
|
184 analysis. This allows one to keep their *.pcl file intact while
|
|
|
185 varying analysis. Hopefully, this avoids errors that may be introduced
|
|
|
186 while manipulating the pcl files.
|
|
|
187
|
|
|
188 Here is an example of the contents of a *.read.config file.
|
|
|
189
|
|
|
190 Matrix: Metadata
|
|
|
191 Read_PCL_Columns: Sample2-Sample15
|
|
|
192 Read_PCL_Rows: Age-Height,Weight,Sex,Cohort-Profession
|
|
|
193
|
|
|
194 Matrix: Abundance
|
|
|
195 Read_PCL_Columns: Sample2-Sample15
|
|
|
196 Read_PCL_Rows: Bacteria-Bug100
|
|
|
197
|
|
|
198 The minimal requirement for a MaAsLin .read.config file is as
|
|
|
199 follows. The Matrix: should be specified. Metadata needs to be named
|
|
|
200 "Metadata" for the metadata section and "Abundance" for the abundance
|
|
|
201 section. “Read\_PCL\_Rows:” is used to indicate which rows are data or
|
|
|
202 metadata to be analyzed. Rows can be identified by their metadata/data
|
|
|
203 id. Separate ids by commas. If there is a consecutive group of
|
|
|
204 metadata/data a range of rows can be defined by indicating the first
|
|
|
205 and last id separated by a “-“. If the beginning or ending id is
|
|
|
206 missing surrounding an “–“, the rows are read from the beginning or to
|
|
|
207 the end of the pcl file, respectively.
|
|
|
208
|
|
|
209 A minimal example is shown here:
|
|
|
210
|
|
|
211 Matrix: Metadata
|
|
|
212 Read\_PCL\_Rows: -Weight
|
|
|
213
|
|
|
214 Matrix: Abundance
|
|
|
215 Read_PCL_Rows: Bacteria-
|
|
|
216
|
|
|
217 With this minimal example, the delimiter of the file is assumed to be
|
|
|
218 a tab, all columns are read (since they are not indicated
|
|
|
219 here). Metadata are read as all rows from the beginning of the pcl
|
|
|
220 file (skipping the first Sample ID row) to Weight; all data are read
|
|
|
221 as all rows from Bacteria to the end of the pcl file. This example
|
|
|
222 refers to the default input files given in the MaAsLin download as
|
|
|
223 maaslin_demo2.\*.
|
|
|
224
|
|
|
225 ### 5\. Optionally, create your R script.
|
|
|
226
|
|
|
227 The R script is used to add code that manipulates your data before
|
|
|
228 analysis, and for manipulating the multifactoral analysis figure. A
|
|
|
229 default “*.R” script is available with the default MaAsLin project at
|
|
|
230 maaslin/input/maaslin_demo2.R. This is an expert option and should
|
|
|
231 only be used by someone very comfortable with the R language.
|
|
|
232
|
|
|
233 ###6. Optional step if using the sfle analysis pipeline. Place .pcl, .read.config, and optional .R files in `../sfle/input/maasalin/input`
|
|
|
234
|
|
|
235 ###7. Run.
|
|
|
236
|
|
|
237 By running the commandline script:
|
|
|
238 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 (src/transpose.py). The following example will have such a call included. An example call from the Maaslin folder for the demo data could be as follows.
|
|
|
239
|
|
|
240 ./src/transpose.py < input/maaslin_demo2.pcl > maaslin_demo2.tsv
|
|
|
241 ./src/Maaslin.R -i input/maaslin_demo2.read.config demo.text maaslin_demo2.tsv
|
|
|
242
|
|
|
243 When using sfle:
|
|
|
244 Go to ../sfle and type the following: scons output/maaslin
|
|
|
245
|
|
|
246 ###8. Discover amazing associations in your results!
|
|
|
247
|
|
|
248
|
|
|
249 #G. Expected Output Files
|
|
|
250
|
|
|
251 The following files will be generated per MaAsLin run. In the
|
|
|
252 following listing the term projectname refers to what you named your "\*.pcl" file without the extension.
|
|
|
253
|
|
|
254 ###Output files that are always created:
|
|
|
255
|
|
|
256 **projectname_log.txt**
|
|
|
257
|
|
|
258 This file contains the detail for the statistical engine. This can be
|
|
|
259 useful for detailed troubleshooting.
|
|
|
260
|
|
|
261 **projectname-metadata.txt**
|
|
|
262
|
|
|
263 Each metadata will have a file of associations. Any associations
|
|
|
264 indicated to be performed after initial variable selection (boosting)
|
|
|
265 is recorded here. Included are the information from the final general
|
|
|
266 linear model (performed after the boosting) and the FDR corrected
|
|
|
267 p-value (q-value). Can be opened as a text file or spreadsheet.
|
|
|
268
|
|
|
269 **projectname-metadata.pdf**
|
|
|
270
|
|
|
271 Any association that had a q-value less than or equal to the given
|
|
|
272 significance threshold will be plotted here (default is 0.25; can be
|
|
|
273 changed using the commandline argument -d). If this file does not
|
|
|
274 exist, the projectname-metadata.txt should not have an entry that is
|
|
|
275 less than or equal to the threshold. Factor data is plotted as
|
|
|
276 knotched box plots; continuous data is plotted as a scatter plot with
|
|
|
277 a line of best fit. Two plots are given for MaAslin Methodology; the
|
|
|
278 left being a raw data plot, the right being a corresponding partial
|
|
|
279 residual plot.
|
|
|
280
|
|
|
281 **projectname.pdf**
|
|
|
282
|
|
|
283 Contains the biplot visualization. This visualization is presented as a build and can be affected by modifications in the R.script or by using commandline.
|
|
|
284
|
|
|
285 **projectname.txt**
|
|
|
286
|
|
|
287 A collection of all entries in the projectname-metadata.pdf. Can be
|
|
|
288 opened as a text file or spreadsheet.
|
|
|
289
|
|
|
290 ###Additional troubleshooting files when the commandline:
|
|
|
291
|
|
|
292 **data.tsv**
|
|
|
293
|
|
|
294 The data matrix that was read in (transposed). Useful for making sure
|
|
|
295 the correct data was read in.
|
|
|
296
|
|
|
297 **data.read.config**
|
|
|
298
|
|
|
299 Can be used to read in the data.tsv.
|
|
|
300
|
|
|
301 **metadata.tsv**
|
|
|
302
|
|
|
303 The metadata that was read in (transposed). Useful for making sure the
|
|
|
304 correct metadata was read in.
|
|
|
305
|
|
|
306 **metadata.read.config**
|
|
|
307
|
|
|
308 Can be used to read in the data.tsv.
|
|
|
309
|
|
|
310 **read_merged.tsv**
|
|
|
311
|
|
|
312 The data and metadata merged (transposed). Useful for making sure the
|
|
|
313 merging occurred correctly.
|
|
|
314
|
|
|
315 **read_merged.read.config**
|
|
|
316
|
|
|
317 Can be used to read in the read_merged.tsv.
|
|
|
318
|
|
|
319 **read_cleaned.tsv**
|
|
|
320
|
|
|
321 The data read in, merged, and then cleaned. After this process the
|
|
|
322 data is written to this file for reference if needed.
|
|
|
323
|
|
|
324 **read_cleaned.read.config**
|
|
|
325
|
|
|
326 Can be used to read in read_cleaned.tsv.
|
|
|
327
|
|
|
328 **ProcessQC.txt**
|
|
|
329
|
|
|
330 Contains quality control for the MaAsLin analysis. This includes
|
|
|
331 information on the magnitude of outlier removal.
|
|
|
332
|
|
|
333 **Run_Parameters.txt**
|
|
|
334 Contains an account of all the options used when running MaAsLin so the exact methodology can be recreated if needed.
|
|
|
335
|
|
|
336 #H. Other Analysis Flows
|
|
|
337
|
|
|
338 ###1. All verses All
|
|
|
339 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 (placed in your args file if using sfle, otherwise added in the commandline call):
|
|
|
340
|
|
|
341 > -a -F cage
|
|
|
342
|
|
|
343 -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:
|
|
|
344
|
|
|
345 > -a -F cage -s none
|
|
|
346
|
|
|
347 #I. Troubleshooting
|
|
|
348
|
|
|
349 ###1\. (Only valid if using Sfle) ImportError: No module named sfle
|
|
|
350
|
|
|
351 When using the command "scons output/maaslin/..." to run my projects I
|
|
|
352 get the message:
|
|
|
353
|
|
|
354 ImportError: No module named sfle:
|
|
|
355 File "/home/user/sfle/SConstruct", line 2:
|
|
|
356 import sfle
|
|
|
357
|
|
|
358 **Solution:** You need to update your path. On a linux or MacOS terminal
|
|
|
359 in the sfle directory type the following.
|
|
|
360
|
|
|
361 export PATH=/usr/local/bin:`pwd`/src:$PATH
|
|
|
362 export PYTHONPATH=$PATH
|
|
|
363
|
|
|
364
|
|
|
365 ###2\. When trying to run a script I am told I do not have permission
|
|
|
366 even though file permissions have been set for myself.
|
|
|
367
|
|
|
368 **Solution:** Most likely, you need to set the main MaAsLin script
|
|
|
369 (Maaslin.R) to executable.
|
|
|
370
|
|
|
371 #J. Installation as an Automated Pipeline
|
|
|
372
|
|
|
373 SflE (pronounced souffle), is a framework for automation and
|
|
|
374 parallelization on a multiprocessor machine. MaAsLin has been
|
|
|
375 developed to be compatible with this framework. More information can
|
|
|
376 be found at
|
|
|
377 [http://huttenhower.sph.harvard.edu/sfle](http://huttenhower.sph.harvard.edu/sfle). If
|
|
|
378 interested in installing MaAsLin in a SflE environment. After
|
|
|
379 installing SflE, download or move the complete maaslin directory into
|
|
|
380 `sfle/input`. After setting up, one places all maaslin input files in
|
|
|
381 `sfle/input/maaslin/input`. To run the automated pipeline and analyze
|
|
|
382 all files in the `sfle/input/maaslin/input` directory, type: `scons output/maaslin`
|
|
|
383 in a terminal in the sfle directory. This will produce
|
|
|
384 output in the `sfle/output/maaslin` directory.
|
|
|
385
|
|
|
386 #K. Commandline Options (Modifying Process and Figures)
|
|
|
387
|
|
|
388 Although we recommend the use of default options, commandline
|
|
|
389 arguments exist to modify both MaAsLin methodology and figures. To see
|
|
|
390 an up-to-date listing of argument usage, in a terminal in the
|
|
|
391 `maaslin/src` directory type `./Maaslin.R -h`.
|
|
|
392
|
|
|
393 An additional input file (the args file) can be used to apply
|
|
|
394 commandline arguments to a MaAsLin run. This is useful when using
|
|
|
395 MaAsLin as an automated pipeline (using SflE) and is a way to document
|
|
|
396 what commandline are used for different projects. The args file should
|
|
|
397 be named the same as the *.pcl file except using the extension .args
|
|
|
398 . This file should be placed in the `maaslin/input` directory with the
|
|
|
399 other matching project input files. In this file please have one line
|
|
|
400 of arguments and values (if needed; some arguments are logical flags
|
|
|
401 and do not require a value), each separated by a space. The contents
|
|
|
402 of this file will be directly added to the commandline call for
|
|
|
403 Maaslin.R. An example of the contents of an args file is given here.
|
|
|
404
|
|
|
405 **Example.args:**
|
|
|
406
|
|
|
407 -v DEBUG -d 0.1 -b 5
|
|
|
408
|
|
|
409 In this example MaAsLin is modified to produce verbose output for
|
|
|
410 debugging (-v DEBUG), to change the threshold for making pdfs to a
|
|
|
411 q-value equal to or less than 0.1 (-d 0.1), and to plot
|
|
|
412 5 data (bug) features in the biplot (-b 5).
|
|
|
413
|
