Optimus: a workflow for LC-MS feature analysis and spatial mapping

Table of contents

• Introduction
• Who might need this workflow?
• What it does?
• System requirements
• Installation
  • Express installation (>= Windows 7 or >= macOS 10.10)
  • Regular installation
  • Installing and updating workflow
• Input
• IMPORTANT: Stub input file
• Experimental design file
• Basic use-case
• Output
• KNIME Basics
• Demo
• Advanced use-cases
• Publications
• Troubleshooting
• License

Introduction

Optimus is a workflow for LC-MS-based untargeted metabolomics. It can be used for feature detection, quantification, filtering (e.g. removing background features), annotation, normalization and, finally, for spatial mapping of detected molecular features in 2D and 3D using the `ili app. Optimus employes the state-of-the-art LC-MS feature detection and quantification algorithms by OpenMS which are joined into a handy pipeline with a modern workflow management software KNIME with additional features implemented by us.

The workflow is being developed by Alexandrov Team at EMBL Heidelberg (contact information) in collaboration with the Dorrestein Lab at UCSD.

• Developer: Ivan Protsyuk
• Principal investigator: Theodore Alexandrov

Who needs this workflow?

The workflow was initially developed for LC-MS-based metabolite cartography, but can be useful in almost any study of LC-MS-based untargeted metabolomics. Direct-infusion experimental data is also supported. Optimus is developed to be open-source, sharable, and efficient enough to process hundreds of LC-MS runs in reasonable time.

What it does?

The workflow consists of the following consequtive steps:

1. Detection of LC-MS features in each run.
2. Alignment and quantification of features detected across all the runs.
3. (Optional) Visualization of the timeline of internal standards intensities.
4. (Optional) Exclusion of features that came from blank/control runs.
5. (Optional) Exclusion of rarely observed features, i.e. features that occur in a small number of runs only.
6. (Optional) Exclusion of features for which MS/MS spectra were not acquired.
7. (Optional) Exclusion of features that are not reproducible in pooled quality control (QC) runs.
8. (Optional) Exclusion of features that are not reproducible in replicate runs.
9. (Optional) Exclusion of features corresponding to compounds eluting in the beginning or end of LC-runs.
10. (Optional) Putative molecular annotation of detected features by mz-RT matching to a list of molecules of interest. This implements a molecular identification at the level putatively annotated compounds, corresponding to the level 2 of the Metabolomics Standards Initiative; see Sumner et al. (2007) Metabolomics, 3(3), 211-221 for details. Note that MS/MS validation of putative annotations is needed (currently not provided in Optimus). The list of molecules of interest can be directly exported from GNPS as a result of MS/MS matching against spectral libraries available at GNPS. Alternatively, the list can be provided as a CSV file created manually.
11. (Optional) Normalization of intensities of detected features. Currently, several normalization methods are available, based on:
    • total ion current (TIC) of each run;
    • internal standards present in all runs;
    • features detected in pooled QC samples.
12. (Optional) Exporting spectral data of detected features for further downstream analysis with third-party tools, including those performing In-Silico prediction such as Sirius or MS-FINDER.
13. Creating a heat map of intensities of detected features across all samples.
14. Visualizing samples on a 3D PCA plot.
15. Creating spatial maps of detected features that can be visualized in `ili app. It is a web-application for interactive visualization of spatial data mapped either on an image or a 3D model, also developed by Alexandrov Team.

System requirements

• Operating system: only 64-bit systems are supported; MS Windows, Linux or Apple macOS.
• RAM: 2 GB is minimal amount. Generally, it is not enough for analysis of large datasets containing about a hundred or more LC-MS runs. However, it is sufficient for smaller ones. After all, it very much depends on the data itself (instrument, mass resolution, LC run-time, etc) and workflow settings.
• CPU: no special requirements. The more powerful your CPU is, the faster Optimus works. It leverages effectively multicore processors for parallel execution which improves the overall performance dramatically.
• Hard drive: all Optimus components take about 2.5 GB. During the execution Optimus creates temporary files that can occupy up to few times more space than initial dataset. Those files are not deleted automatically to enable iterative execution and re-execution of Optimus. However, there is an option inside the workflow to clean up temporary files.

Installation

The workflow is performed by KNIME Analytics Platform, an open-source cross-platform general-purpose workflow management system. Before you start using the workflow, you need to install KNIME itself, Python 2.7 (if it's not already installed) and a few additional modules for Python and KNIME. The installation steps are described below. If your computer is running Windows 7 or newer or macOS 10.10 or newer, you can avail express installation scripts described in the section below. Otherwise, you will need to install Optimus dependencies manually as per Regular installation section.

Express installation (>= Windows 7 or >= macOS 10.10)

Go to the Releases section of this repository, download a zip archive with the latest Optimus version and unpack it to any directory on your computer. Then, follow the instruction for your OS:

Windows users: open the installer subdirectory and double-click win_installer.cmd. It should install KNIME and Python automatically. During the installation, you will be prompted to select KNIME installation directory via a graphical window.

macOS Users: open your Terminal, navigate to the installer subdirectory and execute sudo bash mac_installer.sh.

All: After the installation has finished, make sure that the installed Python distribution is recognized correctly by KNIME:

1. Launch KNIME, click File menu and select Preferences. A Preferences dialog should appear.
2. In the sidebar of the dialog, expand the KNIME item and select its Python sub-item.
3. Click Apply.
4. If no error messages are shown in the dialog, Python distribution was recognized by KNIME successfully, and no further actions are needed. Otherwise, proceed with the following steps.
5. Click Browse... and select file
   • Windows users: C:\Users\<user name>\AppData\Local\OptimusAnaconda\python
   • macOS users: /Users/<user name>/OptimusMiniconda/bin/python
6. Click OK. If no errors are displayed, you can proceed to Installing and updating workflow section.

Regular installation

1. Download and install Python 2.7 64-bit if you don't have installed (it can be the case on Windows). You can download it from the official Python Downloads Page.
   • You can check easily if the needed Python distribution is already installed by typing python -V in your command prompt. You expect to see the output line starting with "Python 2.7". The second part of the version check is determining a bit version of your python interpreter. Follow this instruction to know whether you have a 64-bit Python or not.
   • Windows users: Python installation directory might not be included to your Path environment variable. That's why you might get an error message upon executing python in command prompt although it's installed. To fix this, you should add <Python_installation_directory> to Path as well as <Python_installation_directory\Scripts>. By default, these directories are C:\Python27 and C:\Python27\Scripts. You can find an instruction on changing Path variable here.
   • If you know you have a Python distribution installed, but find steps above too complicated, just install another Python interpreter of the needed version. Multiple versions can co-exist on the same computer fine. However, remember the directory where you install it.
2. Install a few Python modules needed for interaction between KNIME and Python. You can do this by executing the command below in your command prompt.
   • Windows users: Before, you should download and install a Microsoft C++ Complier. One of those Python modules depends on it.
   • pip install six pandas protobuf pymspec pyopenms
   • Others: Before, make sure you have pip package manager available on your workstation. If you don't, execute sudo easy_install pip in the terminal to install it.
   • sudo pip install six pandas protobuf pymspec pyopenms
3. Download and install KNIME Analytics Platform 64-bit. Select a package according to your operating system on the official KNIME Downloads Page.
   • Note: If you already have KNIME installed, make sure that its version is not older than 2.12.*. The workflow hasn't been tested with older versions.
4. Launch KNIME and install additional extensions.
   1. Go to File => Install KNIME Extensions.... Available software dialog should open after this.
   2. Check the following items in the list of available extensions:
      • OpenMS
      • KNIME Python Integration
      • KNIME Quick Forms (legacy)
      • KNIME Virtual Nodes
      • KNIME JavaScript Views
   3. Proceed with the installation. You'll be asked to accept the terms of GPL license at the end.
   4. Restart KNIME after the installation is finished.

Note, that the procedure described above should be completed only once. So, if you get a new version of the workflow in the future, all you'll have to do is just open it with KNIME. As soon as the steps above are accomplished, your environment is ready to run the workflow.

Possible Python compatibility issues: If you have several Python installations in your system, please make sure that KNIME detected the correct one. To do this, go to File => Preferences, then type "python" in the filter box. You should see two items at the left-hand side of the dialog: KNIME > Python. Click at Python and check that there're no error messages appear. If there're any, press Browse... and navigate to the python executable that was called when installing modules at the 3rd step. If you followed the instructions above precisely, you can get a path to the needed python executable by executing which python in Linux/macOS terminal or where python in Windows command prompt.

Installing and updating workflow

It's assumed that you've downloaded the latest Optimus release and extracted it to a directory on your computer.

1. Launch KNIME, go to File => Import KNIME Workflow.... Workflow Import Selection dialog should open after this.
2. Check File, press Browse... and select the Optimus v...knwf file from the extracted directory.
3. Press Finish.

Now, you should see the Optimus v... item in the list at the left-hand side of the KNIME window. Double-click on it to open the workflow in the Workflow Editor where you can change its settings and specify input/output files.

New versions of the workflow appear as new releases in this repository. In order to update the workflow on your local computer to a newer version, repeat the steps above.

Input

The workflow supports mzML and mzXML formats of mass spectrometry data. Internally, all input files will be converted to mzML, so you can save some time on the workflow execution if your data are already in this format. Make sure that input files contain centroided data.

IMPORTANT: Stub input file

The policy of KNIME input nodes implies they always have some files selected. However, it doesn't always match use-cases Optimus can handle, e.g. you might not have a list of internal standards spiked in your samples. To bypass this restriction and make the workflow run, you need to create a file called "stub.txt" anywhere on your computer and use it as an input file whenever you don't have files required by an input node.

Experimental design file

The first stage of Optimus execution is the creation of a file with some details concerning your experimental design such as blank runs, replicate runs, etc. This information can be used by Optimus during the data analysis to remove features caused by background signals or those that are not reproducible in replicate runs. The experimental design is a CSV spreadsheet consisting of 4 columns: file path, LC-run type, sample group and replicate group. Optimus will generate a template of the spreadsheet with all file paths filled, but other columns should be filled manually according to your study as described below:

• LC-run type: BLANK in rows corresponding to blank LC-runs, POOLED_QC for pooled QC runs.
• Sample group: identifiers of your study groups, which can be any text. The column should be filled for all rows or empty. A single LC-run can be a member of several groups. In this case, group identifiers should be separated by semicolon.
• Replicate group: identifiers of a source sample represented by replicates. It can be any text. Every identifier should be used at least twice in the column.

Basic use-case

1. Open the workflow in the KNIME Workflow Editor.
2. Do a right click on the Read LC-MS runs node and select Configure.... A dialog for input file selection should show up.
3. Press Clear, then press Add and select files with your samples and press OK.
4. Create the stub file.
5. Select the stub file as input for nodes Read group mapping, Read list of internal standards and Read annotation source.
6. Right-click on Where to save template of experimental design and select Configure.... A dialog for output directory selection should show up.
7. Click Browse... and specify where a template file with your experimental design will be stored. It's recommended to keep it in the same directory where you LC-MS data is located. Click OK.
8. Right-click on the Generate template of experimental design node and select Execute. The upper part of the workflow should start execution, and the file with experimental design will be created. The file can be then edited manually according to your experimental design as described above.
9. Right-click on the Read experimental design node and select Configure.... A file selection dialog should appear.
10. Click Browse... and select the experimental design file.
11. Make sure that read column headers is checked, read row IDs is not checked, and column delimiter is set to , (comma). Click OK.
12. Right-click on the Display feature heat map node and select Execute. The workflow should start execution. It's finished when a red circle in the lower part of the node turns into a green one. Wait till it happens.
13. Right-click on the Display feature heat map node and select Interactive View: Generic JavaScript View. A window showing distribution of detected features will show up.
    • Note: by default, log transformation is applied to intensity values before rendering them on the heatmap. You can switch to initial intensities using Scale at the left-bottom corner of the window.

Output

In order to save results produced by Optimus, open the configuration dialog of the Save results node and specify an output directory. Once the node is executed, it creates 3 files in the output folder. One of them, features_quantification_matrix.csv can be opened in any spreadsheet editor (e.g. Excel). Rows in the table will correspond to input samples, whereas columns will represent consensus features, i.e. ions of the same type quantified across the runs. Table cells contain intensities of corresponding features. Names of columns give information on corresponding features. The format is mz_value RT charge (ID: numeric_identifier), so for example a column named 233.112 69 1 (ID: 123) represents a single-charged ion with mz-value about 233.112 and chromatographic peak at around 69 seconds. As reported features are consensus, it doesn't mean one can find a mass trace in input samples matching consensus mz-value and retention time exactly. These figures are averaged across the runs, but the variation is supposed to be low from run to run.

Numeric identifiers (IDs) are assigned to features after the alignment step and are not changed at the further steps. For the same input dataset and fixed parameters of feature detection and alignment, association between IDs and features are guaranteed to remain the same. So, the IDs can be used as shortcuts for features.

Another file produced by the workflow, Optimus_settings.ini is a configuration file that contains the list of values of all Optimus parameters used to generate the output. This file, along with the experimental design, can be used to reproduce the data analysis.

The third OptimusViewer_input.db file contains extracted ion chromatograms (XIC) and MS/MS spectra for detected features. The file can be opened with OptimusViewer application also developed by Alexandrov team. You can download it and find the instruction on usage in this GitHub repository.

KNIME Basics

If you're new to workflow management systems or KNIME in particular, you can find an introductory tutorial on basic features of KNIME here.

Demo

This repository contains real-life samples that you can test the workflow on. They're available in this archive (courtesy of Alexey Melnik, Dorrestein Lab, UCSD). Inside, you'll find a directory called samples that contains LC-MS samples in mzXML format ready to be processed with the workflow. Blank samples separated from the normal ones in the blanks directory inside samples. They can be used to remove background features from your result features set. There're also 2 files in the root folder called coords.csv and Rotten_Apple_Model.stl. You'll need to supply the former one at the last step of the workflow that is supposed to produce spatial maps for `ili.

If you want to check quickly, what are actually the results of the workflow, without diving into KNIME and installing everything, you can find the needed file in the results folder in the archive. It contains file features_mapping.csv which is a spreadsheet containing a table with intensities of different features detected in different runs. This file can be visualized in `ili along with Rotten_Apple_Model.stl. You can simply drag&drop both of them to the `ili window.

Below, you can find an example of a spatial map obtained from `ili for a feature that is localized mainly in the vicinity of rot on the apple.

Advanced use-cases

The workflow has many capabilities that you can discover in the documentation embedded into it. Click on any node, and the description of its role and its parameters will show up in the banner at the right-hand side of the KNIME window. Different nodes don't depend on each other, so you can experiment with different settings and track changes of the workflow output.

Publications

• Ivan Protsyuk, Alexey V Melnik, Louis-Felix Nothias, Luca Rappez, Prasad Phapale, Alexander A Aksenov, Amina Bouslimani, Sergey Ryazanov, Pieter C Dorrestein & Theodore Alexandrov. 3D molecular cartography using LC–MS facilitated by Optimus and 'ili software. Nature Protocols, 2018. doi (Please cite this paper to acknowledge Optimus in your publication)
• Alexey V. Melnik, Pieter C. Dorrestein, Ricardo R. da Silva, Embriette R. Hyde, Alexander A. Aksenov, Fernando Vargas, Amina Bouslimani, Ivan Protsyuk, Alan Jarmusch, Anupriya Tripathi, Theodore Alexandrov, and Rob Knight. Coupling targeted and untargeted mass spectrometry for metabolome-microbiome-wide association studies of human fecal samples. Analytical Chemistry, 2017. doi
• Daniel Petras, Louis-Félix Nothias, Robert A. Quinn, Theodore Alexandrov, Nuno Bandeira, Amina Bouslimani, Gabriel Castro-Falcón, Liangyu Chen, Tam Dang, Dimitrios J. Floros, Vivian Hook, Neha Garg, Nicole Hoffner, Yike Jiang, Clifford A. Kapono, Irina Koester, Rob Knight, Christopher A. Leber, Tie-Jun Ling, Tal Luzzatto-Knaan, Laura-Isobel McCall, Aaron P. McGrath, Michael J. Meehan, Jonathan K. Merritt, Robert H. Mills, Jamie Morton, Sonia Podvin, Ivan Protsyuk, Trevor Purdy, Kendall Satterfield, Stephen Searles, Sahil Shah, Sarah Shires, Dana Steffen, Margot White, Jelena Todoric, Robert Tuttle, Aneta Wojnicz, Valerie Sapp, Fernando Vargas, Jin Yang, Chao Zhang, and Pieter C. Dorrestein. Mass Spectrometry-Based Visualization of Molecules Associated with Human Habitats. Analytical Chemistry, 2016. doi

Troubleshooting

Some errors can appear in the application log that interrupt workflow execution. A node caused an error will be the left-most node with a red circle in its right side. In addition, you should see error output in KNIME Console. Below, you can find solutions for some common issues.

License

The content of this project is licensed under the Apache 2.0 licence, see LICENSE.md.