This is a stand-alone implementation of the new updated PRIDE Cluster algorithm. It is based on the spectra-cluster API and uses a highly similar logic as the Hadoop implementation spectra-cluster-hadoop used to build the PRIDE Cluster resource.
You can find an overview over all our clustering related tools at https://spectra-cluster.github.io.
WARNING: This software is still in beta phase. We do expect it to still have several bugs. Bug reports are highly welcome! Just submit in issue to let us know.
- Fixed a bug that caused a crash when re-binning very small datasets with scarce m/z regions.
- Fixed a bug when running on systems with non-english locale
- Fixed a bug in the merging of adjacent bins: Spectra were not re-binned correctly
- Adapted binning procedure to create fewer files
- Updated to spectra-cluster API version 1.1 Therefore, the following new features are available
- Automatically detect the number of comparison to use. This no longer has to be set by the user.
- A new command line option was added (
-x_min_adapt_comparison) that sets a minimum number of comparisons that will always be used (as a correction factor) in case the number learned from the data is smaller. This feature is only recommended when clustering very small datasets.
- New command line option
-add_scoreswas added. This then outputs additional scores to the generated .clustering file. These additional properties are stored as JSON encoded strings. - Added command line options to only cluster identified or only cluster unidentified spectra.
- For a complete list of command line options, please refer to the output of the
-helpoption
- Updated to spectra-cluster API version 1.0.11. Changes include:
- New version of the .clustering file format which stores how often a consensus spectrum peak was observed.
- Option to add debug information to spectra such as the score when the spectrum was merged to a cluster.
- Added support for incremental clustering. The spectra-cluster-cli tool can now process MGF files and .clustering files. This only works with .clustering files created with the spectra-cluster-cli tool version >= 1.0.4.
- Changed to spectra-cluster API version 1.0.10 which includes contaminant peak filters
- Added CLI options to remove contaminant peaks from spectra.
- Changed to spectra-cluster API version 1.0.9. This fixes issues caused by unknown charge states (as well as #2).
- updated to new .clustering format version
- .clustering files now includes complete reference to original spectrum using the same indexing system as the PSI standard file formats
- added feature to learn the cumulative distribution function (CDF) from a given dataset and then use this newly learned CDF
- added a list of experimental / advanced parameters (all starting with "-x_"). For more information simply launch the application with the "-help" parameter
- Note: For small datasets (<100 MS runs), the option
-x_min_comparisonsshould be set to 10,000. - fixed #3 see 1db57ae
The spectra-cluster-cli application is written in Java and therefore runs on Windows, Linux, and Mac OS X.
Java: needs to be installed on your system for the spectra-cluster-cli to work. You can download the latest Java version for your system here.
To install the spectra-cluster-cli simply download the latest release and extract the zip file.
The spectra-cluster-cli tool is a command-line only tool. If you prefer to use a graphical use interface, please use the spectra-cluster-gui tool. A detailed tutorial on how to prepare your files for clustering can be found at https://spectra-cluster.github.io.
To use the spectra-cluster-cli tool, follow these steps:
Open a command line and navigate to the folder where you extracted the spectra-cluster-cli to. The following example assumes that you are already in this folder.
This command launches the clustering job using the default values (as used for the PRIDE Cluster resource) all available CPU cores and writes the results to my_clustering_result.clustering.
Note: You need to replace the spectra-cluster-cli-1.1.0.jar with the name of the downloaded version.
$ java -jar spectra-cluster-cli-1.1.0.jar -filter immonium_ions -output_path my_clustering_result.clustering C:\my_first_file.mgf C:\my_second_file.mgfTo improve the clustering accuracy in small datasets (< 100 MS runs) the default value for -x_min_comparisons was changed to 10,000 (changed in version 1.0.3). When clustering a repository scale dataset, a value of 5,000 is used (default value in the Hadoop version).
Additionally, we recommend to always either use the immonium_ions filter, or even filter all peaks below 150 m/z ("-filter mz_150") or even below 200 m/z ("-filter mz_200").
The full list of options is printed through the -help parameter:
$ java -jar spectra-cluster-cli-1.1.0.jar -helpThe spectra-cluster-cli generates a .clustering file to store the clustering results. A specification of this format can be found at the clustering-file-reader page
The spectra-cluster-py Python library contains a collection of tools to analyse the clustering results. You can find a detailed documentation of this library at http://spectra-cluster-py.readthedocs.io. Additionally, this library contains many classes that should help in writing your own Python scripts to analyse your clustering results.
Additionally, we provide a Java API that can be used to develop Java software that reads the .clustering file format.
Should you have any problems when running the spectra-cluster-cli tool, please do not hesitate to report this problem using the issue tracker.
In case you have any other questions don't hesitate to post a question at http://qa.proteomics-academy.org.
These can include
- questions about what to do with the results
- questions about whether the clustering algorithm can be used for a given analysis problem
We are using the Hadoop version version of the spectra-cluster algorithm to cluster the complete PRIDE Archive repository. Thereby, we were able to recognize millions of consistently unidentified spectra across thousands of submission. These clustering results are presented in the PRIDE Cluster resource.
For more information see the recent paper Griss et al., Recognizing millions of consistently unidentified spectra across hundreds of shotgun proteomics datasets., Nat. Meth. 2016 Aug;13(8):651-656.
Additionally, if you are able to use our algorithm for your own project, please cite the above reference.