Home

News

06/11/2024

Enjoy the Zero-install version of Prostar as well as the new Online demo!

  • Prostar version 1.38.0 has been released on the Bioconductor (release 3.20) and is deployed as a Zero-install zip.

16/01/2024

Meet us at Winterberg for EuBICS Winter school! (https://eubic-ms.org/events/2024-winter-school/).

To download for the Prostar workshop:

  • Download the zero install version of Prostar (Windows only) Prostar 1.36.3 zip

  • Or run it through Docker with the following command:

    docker run -it -p 80:3838 ghcr.io/edyp-lab/prostar:1.38.0

    Then, access it by the following link : Prostar

  • Alternatively, if you have R already installed, Prostar can be installed and launched with the following commands:

    pkgs <- c('edyp-lab/DAPARdata', 'edyp-lab/DAPAR', 'edyp-lab/Prostar')
    install.packages('remotes')
    remotes::install_github(pkgs, dependencies = TRUE, ref = RELEASE_3_20)
    Prostar::Prostar()

It takes about 10 minutes to install Prostar and it dependencies from a new R installation

  • Several datasets will be used, they can be downloaded here.

About

Prostar is a software tool dedicated to the differential analysis of quantitative data resulting from discovery proteomics experiments.

Prostar is easy to install (see our Zero-install page), easy to use (thanks to its Shiny-based click-button interface) and well-documented (see our reference page). Moreover, it has been regularly updated along years to provide state-of-the-art data science methodologies.

Citation

Maintaining Prostar as free software is a heavy duty. Please cite the following reference

S. Wieczorek, F. Combes, C. Lazar, Q. Giai-Gianetto, L. Gatto, A. Dorffer, A.-M. Hesse, Y. Coute;, M. Ferro, C. Bruley and T. Burger. DAPAR & ProStaR: software to perform statistical analyses in quantitative discovery proteomics. Bioinformatics 33(1), 135-136, 2017.
http://doi.org/10.1093/bioinformatics/btw580.

Presentation

Data management

  • Conversion: To import a tabulated file containing quantitative data and convert it into an MSnset structure.
  • Loading: To open an Msnset structure that has been previously constructed.
  • Exporting: To save a partially/completely processed dataset and to download the data analysis results.
  • Demo data: Toy datasets are available to discover Prostar potential in the simplest way.

Data processing

  • Filtering: To prune the protein or peptide list according to various criteria (missing values, string matching).
  • Normalization: To correct batch or group effects.
  • Imputation: By taking into account the very nature of each missing value.
  • Aggregation: For peptide-level datasets, it is possible to estimate protein abundances.
  • Hypothesis testing: To compute the significance of each protein differential abundance.

Data mining

  • Descriptive statistics: Available at any stage of the analysis, for data exploration and visualization.
  • Peptide-Protein Graph: Explore and visualize peptide-protein graphs.
  • Differential analysis: To select a list of differentially abundant proteins with a controlled false discovery rate.
  • Gene Ontology analysis: To map a protein list onto GO terms and test category enrichment.

Software

  1. Zero-install
    The easiest way. Prostar is deployed either as a zip archive (so far only available on Microsoft Windows desktop machines) or as a Docker image (bêta, but ok for Windows, Mac OS X and (Unix/Linux).
  2. Stand-alone Bioconductor install
    The standard method to install Bioconductor distributed software. This method works for any operating systems (Unix/Linux, Mac OS X and Windows) as long as R is installed.
  3. Server install
    When one wants Prostar to run on a Unix server, on which remote users connect. This more advanced type of install is detailed in the user manual.
  4. Online demo
    Before installing Prostar on your desktop machine, test our online demo!

Zero-install

Zip file (only for Windows)

Remarks: 1. It is not necessary to already have R installed.
2. For now, Zero-install is only available for Microsoft Windows machines.
3. At first launch, an internet connection is necessary to finish the install.
4. To ensure full compatibility and debugging, the zip file may be available up to few weeks after each Bioconductor release.

Just download the zip file below and unzip it! The unzipped folder contains an executable file which directly launches Prostar.

Download Prostar 1.38.0 zip file (Release date: 06/11/2024)

Download Prostar4metabolomics 1.22.8 zip file (Release date: 22/06/2021)

Bêta Docker image (ok for Unix/Linux, Mac OS X and Windows)

This feature is deployed as a Beta-test, as to give an alternative to the zip files. The Docker image and the zip file are synchronized and embed identical Prostar versions (to ensure full compatibility and debugging, the zip file and the docker image may be available up to few weeks after each Bioconductor release).

As a prerequisite to run the Docker image, Docker must be installed on the computer and the service must be started. The Docker image of Prostar is available on Prostar proteomics’s Github webpage. It can be run using the following command:

docker run -it -p 80:3838 ghcr.io/edyp-lab/prostar:1.38.0

Then, access it by the following link : Prostar

Bioconductor installs

Only stand-alone install is detailed below. For server install, please refer to the user manual. This type of install works with any operating system among Unix/Linux, Mac OS X and Windows. However, it is necessary to have the latest version of R (>= 4.4.0 for Prostar 1.38.x) installed in a directory where the user has read/write permissions.

  1. Install Bioconductor package manager with the following commands (see this page for details):
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install(version='3.20')
  1. Install Prostar:
BiocManager::install("Prostar")
  1. Launch Prostar:
library(Prostar)
Prostar()

Github installs

pkgs <- c('edyp-lab/DAPARdata', 'edyp-lab/DAPAR', 'edyp-lab/Prostar')
install.packages('remotes')
remotes::install_github(pkgs, dependencies = TRUE, ref = 'RELEASE_3_20')
Prostar::Prostar()

Online demo

An online demo of Prostar software is available at: http://live.prostar-proteomics.org

Remark: The server hosting the online demo has limited capacities. Thus, uploading your own dataset may lead to server overload. To test the online demo, please rely on the toy datasets that are available in “Demo data” (from “Data manager” menu).

Release notes

Roadmap for Prostar 2.0

New features

  • To Be Announced - Many Great stuffs! 😉

News in Prostar 1.38

News in Prostar 1.36

Bugs fixed

[1.36.3] (05/10/2024)

  • Bug fixed in the Convert tool when using the ‘Order by conditions’ feature
  • Bug fixed in the Convert tool when clicking on “Convert Data”: sometimes the following error occured: Error in [.data.frame(tmp.df, , ‘Sample.name’) : undefined columns selected.
  • The page ‘Check for Updates’ is back

Bugs fixed

[1.36.2]

  • Bug fixed in the Convert tool when using the ‘Order by conditions’ feature

Bugs fixed

[1.36.1]

  • Bug fixed in hypothesis test tool with Limma when using a dataset in which the names of the samples contain spaces
  • Differential analysis tool. A warning message has been added if the user selects a p-value threshold that is over the p-values of all the data in the volcanoplot. In this case, the FDR cannot be computed.
  • Peptide-protein graph: bug fixed to view the list of One-One connected components in the table and when sorting elements in the tabs.
  • Bug fixed in the convert tool when the use do not select any column for the identification type of samples.

New features

[1.36.0]

  • Hypothesis testing : it is now possible to change the log(FC) threshold and reprocess the test without resetting the entire interface
  • Filtering tool: as for the cell metadata filtering, when users select infos of symbolic or numerical nature, the number of lines which will be deleted is displayed before clicking on the ‘Perform’ button.
  • Convert dataset: fixed a bug which built wrong cells metadata when using the option ‘Order by conditions’ (in the Build design step).

News in Prostar 1.34.6

Bugs fixed

  • Hypothesis testing: bug fixed with Limma and designs of level 2 and 3.

News in Prostar 1.34.5

Bugs fixed

  • Differential analysis: fixing the details of quantitative values of peptides (when a user clicks on a protein in the volcanoplot).
  • Hypothesis testing: allow the use of Limma with dataset containing up to 26 conditions
  • Bug fixed in the tree for the selection of cells metadata (used in Filtering tool and Descriptive Statistics)

News in Prostar 1.34.2

Bugs fixed

  • Bug fixed while opening the ‘Descriptive statistics’ tab after have used one of the processes in ‘Data processing’ tab.

News in Prostar 1.34.1

New features

  • Allow hypothesis test using Limma with a dataset built on a flat design and containing up to 26 conditions

News in Prostar 1.34.0

Bugs fixed

  • Bug fixed while opening xlsx files.
  • Bug fixed while converting either text nor Excel files
  • Bug fixed with the display of some pages

News in Prostar 1.32.1

New features

  • New selection tool for cell metadata tags (used in the filtering tool, in Descriptive Statistics and Differential Analysis)
  • Export as Excel file: additional sheet with DAPAR and Prostar version number used to process the dataset

News in Prostar 1.30.7

Bugs fixed

  • Buggy plots have been temporarly disabled in aggregation process
  • Bug fixed while converting a text/Excel file to a MSnset file (issue with the versions of Prostar and DAPAR)

News in Prostar 1.30.6

Bugs fixed

  • Bug fixed in agregation tool for peptide-level pipeline

New features

  • A popup windows appears if the dataset which is loaded was built with an older version of Prostar than the running one. Prostar will continue to run but the user is informed that unexpected behaviour may happen in this case.

News in Prostar 1.30.5

Bugs fixed

  • Bug fixed in the filtering tool with the cell metadata tags

News in Prostar 1.30.4

Bugs fixed

  • Small bug fixed in the normalization tool

News in Prostar 1.30.3

Bugs fixed

  • In the imputation tool: all imputation methods are functional again
  • Bug fixed in the Student’s test (the value of logFC is now correct and consistent with Limma)

New features

  • In case of error during the execution of some processes(), a popup window - containing the text of the error - will be shown.

News in Prostar 1.30.2

Bugs fixed

  • In the normalization tool:
    • the intensity plots (violinplot and boxplot) are displayed again
    • the display of the comparative plot (before and after normalization) is faster, which speeds up the overall normalisation process.

News in Prostar 1.30.1

New features

  • Differential analysis : a new information has been included concerning the ‘push p-value’ feature. This information is the number of entities that have been pushed
  • Missing values imputation in pipeline at protein: In the ‘Missing on Entire Condition’ step, the info text after impuation was confusing. Now, when the imputation has been achieved a more explicit message( giving the number of imputed proteins) is displayed
  • Metacell tags have been renamed. For more information, please got to the FAQ

Bugs fixed

  • In the filtering tool, when selecting a tag which not appeared in the dataset, all the lines were deleted. This has been fixed. Moreover, the list of tags to choose now reflects only the tags that are really in the dataset and not all the tags available in DAPAR
  • Bug fixed in “Open a MSnset” (“Attempt to select less than one element in get1index”)

News in Prostar 1.28

New features

  • In the ‘Hypothesis testing’ tool, the interface for the ‘Swap conditions’ option has been updated (a collapsible panel has been added).
  • Textual information has been added in the imputation tool when the dataset contains no missing values. This prevents any user to run an imputation on such datasets (which made some imputation methods crash)

Bugs fixed

  • [1.28.5]
    • Improvements of the Differential Analysis tool, especially the information and tooltips related to the volcanoplot.
  • [1.28.0]
    • The ‘Reset’ action button in ‘Differential Analysis’ has been fixed. Now, the ‘Push p-value’ widgets are also set back to default values.
    • Bug fixed when exporting a peptide dataset as an Excel file
    • The ‘Reload Prostar’ button works now (Prostar used to crash when the user clicked on the button)
    • Bug fixed in ‘Normalization’ tool when normalizing with tags in a ‘Specific column’
    • Bug fixed with the Push p-value feature in ‘Differential analysis’ tool.
    • Prostar can now handle peptide datasets in which the protein accession contains “|” (The problem occurred during connected component computation)
    • Bug fixed about the display of quantitative metadata in the data explorer (the columns and their names were not aligned)
    • Bug fixed about Prostar freeze when saving the aggregated dataset.

News in Prostar 1.26

New features

  • In the ‘Convert’ tool, implementation of a new way to select the nature of quantitative data for each sample
  • Add colors to cells metadata tags in Excel export file
  • The ‘swap volcanolplot’ feature in Differential Analysis has been removed. It is replaced by a new feature in the ‘Hypothesis Test’ tool where it is now possible to swap the conditions of any comparison. This leads to a new list of comparisons which are available in the ‘Differential analysis’ tool.

Bug fixed

  • [1.26.4]
    • Fixes a bug with the Swap feature in the Hypothesis Test tool. When the user has swaped a first comparison and want to swap a second comparison, the first one was automatically swaped again.
    • Fixed a bug with the Reset button in the differential analysis which did not reset the parameters of the Push p-value tool.
  • [1.26.3]
    • The function Prostar() works now when the user wants to launch Prostar via a R console.
  • [1.26.2]
    • The selection of tooltips in the volcanoplot works now. It no longer stay on the default one but is updated with the current user selection for tooltips.
    • Bug fixed with the ‘Reset’ button in the ‘Hypothesis Test’ tool
  • [1.26.1]
    • A bug has been fixed in the computation of the t-test (in the Hypothesis test tool). The logFC value for a given comparison was swaped and not consistent with the computation of LogFC in Limma.
    • In the Differential Analysis tool, the small info table about the number of selected entities is showed
    • A bug fixed in the Filter tool with the tab ‘Quanti. metadata filtering’ or with the ‘Push p-value’ feature in the Differential Analysis tool. Now, both ‘keep’ and ‘delete’ options work.
  • [1.26.0]
    • Bug fixed in demo datasets (package ‘DAPARdata’) related to the cell metadata tags for quantitative values.
    • A bug in the function which exports datasets to MSnSets has been fixed.
    • ‘Aggregation tool’: Bug fixed with the ‘proteinId’ on new aggregated protein dataset.
    • Bug fixed with the feature ‘#/% of values to delete’ when selecting ‘%’. This bug occurred in the ‘Filtering tool > Quanti. metadata filtering’ and the ‘Differentiel analysis > Push p-value’
    • Bug fixed in select metadata information for Aggregation

News in Prostar 1.24

New features

  • Fully operational peptidomics and peptide-level workflows.
  • Addition of a cell metadata tag about quantitative values (it indicates for each condition and each entity (either protein or peptide) whether it is a quantitative, missing or imputed value.
  • The missing value filtering tool has been replaced by a filter tool that operates on quantitative cell metadata. A more user-friendly GUI is provided.
  • Missing values barplots have been generalized to any type of cell metadata tag.
  • The information available in cell metadata is now accounted for during the aggregation step.
  • Better color managment in plots.
  • When converting a dataset into a MSnSet object, the log-transformation is run after 0 values have been replaced by NAs.

Bug fixed

  • [1.24.8]
    • Bug fixed in the Convert data tool (when selecting the identification method of quantitative data). The chechbox widget has been replaced by two radio buttons to improve the GUI experience. When the user choose to select identification methods, it is now mandatory to choose appropriate column names.
    • Add comma as separating character for protIds (Use to generate adjacency matrices)
  • [1.24.7]
    • Bug fixed within the ‘Convert tool’, in the ‘Build design’ tab where the use of ‘Order conditions’ did not apply to the quantitative cell metadata. Now, this functionality works.
    • The ‘push p-value’ feature in the Differential Analysis tool works fine now!
  • [1.24.6]
    • Bug fixed with the ‘Reset’ button in the ‘Hypothesis test’ processing tool.
  • [1.24.5]
    • Bug fixed with download Excel and csv buttons. Now, it is possible to download the entire table rather than only 153 items.
    • All spaces in column names have been replaced by ’_’ to standardize names.
    • A ‘keyId’ field has been in the aggregated dataset. It is the same feature as in the Convert tool but, during the aggregation process, it is set automatically - based on the adjacency matrix)
  • [1.24.4]
    • During the conversion process, the following error (“CreateMSnSet: unused argument (indFData=indexForFData”) has been fixed.
    • The push p-value interface is now consistent with the one in the filtering tool. Furthermore, the user cannot run successive “Push p-value” operations. if he runs several times the “Push p-value button”, the dataset is now automatically reset to the original comparison.
    • Bug fixed in the plot which shows the number of lines with <quant. metadata> tags which appears in the filtering tool (tab “Quanti. metadata filtering”) and the menu “Descriptive statistics” (tab “Quantitative nature”)
    • The error message “Argument ‘obj’ is missing with no default” in the preview filtering example (in the tab “Quantitative nature”) has been fixed.
  • [1.24.3]
    • Update the download functionality to get the list of proteins and peptides that make the aggregation failed.
    • Bugs fixed in plots of ‘Quantitative nature’ of entities
  • [1.24.2]
    • Bug fixed with the Reset button in Differential Analysis, as well as in the ‘Push p-value’ functionality
  • [1.24.1]
    • Bug fixed with the number of digits when displaying numbers (such as in quantitative data tables)
  • [1.24.0]
    • Enhancement of the pi0 value selection in differential analysis tool.
    • Bug fixed when the user wants to impute NAs after the normalization tool display (without process any normalization).
    • Bug fixed in differential analysis: The ‘Push p-value’ function now also works on One-vs-All comparisons.
    • Bug fixed with normalization on some selected proteins.

News in Prostar 1.22

New features

  • Functionalities enabling the focus on a protein (or a subset of proteins) to compare the normalization options.
  • Possibility to apply normalization with respect to a user-defined subset of reference proteins.
  • Missing values filtering: It is now possible to tune the filtering option with proportions, in addition to with absolute values.
  • Missing values filtering and “Push p-values” (in Differential analysis tool): one can now select either numeric values or percentage of NA or imputed values.
  • In DAPAR, implementation of ANOVA tools

Bug fixed

  • [1.22.10]
    • Bug fixed in convert tool
  • [1.22.6]
    • Fixed issue when converting a dataset into MSnSet and select columns which correspond to the origin of quantitative values (e.g. ‘By_MS/MS’, ‘By_Matching’, etc..)
    • Bug fixed in normalization tool when the user wants to normalize on a selection of protein. Now, there is no need to check ‘Synchronize’ option to proceed to the normalization.
    • The volcanoplot in the differential analysis tool now updates automatically when the user clicks on the ‘Push p-value’ button.
  • [1.22.5]
    • Bug fixed when Prostar is used with R > 4.0.0 (Convert tool crashes)
    • MEC imputation crashed after POV imputation using KNN
  • [1.22.4]
    • Bug fixed in Welch/Student hypothesis tests.
  • Bug fixed in the convert tool, introduced by R-4.0.3,
  • The extension of the exported file has been corrected
  • Bug fixed in MEC imputation after the POV imputation proceeded with KNN

News in Prostar 1.20

New features

  • Implementation of for more complex experimental designs,
  • Automatic clustering of protein expression profiles,
  • Additional preliminary filtering option for match between run based evidence.

Bug fixed

  • Typos corrections
  • Optimisation of Convert Data tool

News in Prostar 1.18

New features

  • Warning on the FDR if the number of selected peptides/proteins is too small.
  • Computation of the number of shared and specific peptides per protein in the aggregation tool. Notably useful for filtering.
  • Filters on numerical values.
  • New tool for exploring and visualize peptide-protein graphs.
  • New navigation principle to switch between differents steps of a process.
  • Better default filenames when the user click on downloads Button or on export buttons above the different tables in Prostar

Bug fixed

  • [1.18.6]
    • Due to some instability of cache memory when successively opening several datasets in a Prostar session, data management has been simplified. To work on another dataset than the current one, reloading Prostar first is now necessary (with the button above). It will restart Prostar with a fresh R session where import menus are enabled ‘Dataset manager’ menu.
  • [1.18.4]
    • Spinner wheels showed during the computation of plots were replaced by a progress bar at the bottom right of the window.
  • [1.18.3]
    • Bug fixed in retro-compatibility with certain MSnset datasets created with previsous versions of Prostar.
    • Bug fixed in the Color settings UI.
  • [1.18.2] Bug fixed in the volcanoplots.
  • [1.18.1] Bug fixed in the dataset convert tool.
  • In the convert tool (build design step), the reordering of conditions is now functional. In the same screen, bug fixed when the user wants to show design examples.
  • Export of a dataset when several metadata are selected to be integrated to the export file
  • Bug fixed in the navigation bar for each process: the reset button is functional
  • Inconsistent behaviour when the user changes a dataset in any process UI

News in Prostar 1.16

New features

  • New customized theme according to Prostar logo colors.
  • Uniform organization of the method parameters.
  • Three buttons are displayed above each tab to export the data (Copy to clipboard, export to CSV, print).
  • In the CV distribution plot (Descriptive statistics), a convenient zoom is predefined.

Bug fixed

  • [1.16.10] Bug fixed in imputation tools (package imp4p) when the dataset has not samples grouped by conditions. Bug fixed which appeared sometimes in the Export tool.
  • [1.16.9] Bug fixed in hypothesis testing when the dataset has samples not grouped by conditions
  • [1.16.8] Bug fixed: The logFC value calculated by t.test was the opposite as the real one (-log(FC))
  • [1.16.7] Bug fixed which have made wrong Fold Change computation in a specific case : when the dataset contains conditions with different number of samples AND the names of the conditions were not in an alphabetic order
  • [1.16.6] Delete ‘copy to clipboard’ button in the ‘Bug report’ screen
  • [1.16.5] Bug fixed in intensity values during conversion
  • [1.16.4] Bug fixed when converting into MSnset some datasets with no shared peptides
  • [1.16.2] Set ‘No’ as default value for order conditions in convert data
  • [1.16.1] Colors of different conditions in missing values plots ()
  • Interactive quantitative data display from volcanoplot.
  • Aggregation statistics corrected.
  • Differential analysis is now working with Firefox.
  • Sum aggregation function is back.
  • In string-based filtering, reusing several times the same column is possible.
  • Added more numerical precision of p-values in exported dataset (differential analysis)

News in Prostar 1.14

Bug fixed

  • Auto reset of dropdown menu in differential analysis.
  • In the feature metadata table, the FC tag has been replaced by ‘logFC’.
  • In the experimental design table, the column names ‘Experiment’ and ‘Label’ have been replaced respectively by ‘Sample.name’ and ‘Condition’.
  • Delete the dependency to the package imputeLCMD.
  • Tooltip persistance dealt with.

New features

  • Better managment of dropdown menus in the main menu.
  • Add a Bug report tab in the ‘Help’ menu.
  • Reorganization of the menus into Data preprocessing and Data mining.
  • Add proportions in logFC distribution plot.
  • Add LOESS normalization.
  • Add VSN normalization.
  • Improve automatic report generation.
  • New peptide-to-protein aggregation with fair account of shared peptides.
  • Peptide visualization on protein volcano plots.
  • Add customisation of colors for plots.

News in Prostar 1.12

Bug fixed

  • Normalization: “Sum by columns” has been modified to provide log-abundances compatible with subsequent processing.
  • Normalization: Any normalization can now be applied “for each condition independantly” or “globally”.
  • Imputation: All methods are now only applied “for each condition independantly”.

New features

  • The entire pipeline is now compatible with datasets with more than 2 conditions.
  • Descriptive statistics: The expression datasets are colored with respect to the nature of missing value (either POV or MEC, see below), even when the value has been imputed.
  • Filtering: Manage designs with more than 2 conditions and with conditions containing different number of samples.
  • Filtering: More user friendly interface for the string-based filtering (Tab 2).
  • Imputation (protein level): Distinction between missing values on an entire condition (Missing on the Entire Condition - MEC) and the other ones (Partially Observed Value - POV).
  • Imputation (protein level): for POV, it is possible to use SLSA which take into account the experimental design.
  • Differential analysis: All tests can be applied on datasets with different number of samples in each condition.
  • Differential analysis: Limma takes into account all the hierarchical experimental designs.
  • GO analysis: the GeneID nomenclature is now available.

News in Prostar 1.10

Bug fixed

  • When the aggregation step has been performed, the interface switches to the first tab of the ‘Descriptive Statistics’ in order to view informations aout the new dataset (the protein one).
  • Implementation of a parallel version of the function which saves the (new) protein dataset after the aggregation step.
  • Disable the extra row appearing in the metadata table when convertinga text file to a MSnSet file.
  • A new package (readxl) is used to read xls or xlxs files. In certain circumstances, the functions of the previous package openxlsx is not able to decode properly Excel files.
  • When converting a new (text or Excel) file in Prostar : the missing values were not registered as expected. Especially, they did not appear in blue in the table above the volcanoplot. Bug fixed
  • A bug occured when the user load successively several datasets in Prostar. The previous ones were note correctly erased and this has lead to side effects. This bug is now fixed

New features

  • Gene Ontology (GO) analysis (Beta version).
  • Automatic report generation (Beta).
  • Preliminary separation between peptide and protein level pipelines.
  • IMP4P method for peptide level imputation.
  • DetQuantile method for protein level imputation.
  • Tooltip implementation.
  • Interactive plots with highcharter.
  • Enhancement of the string-based filtering UI

Support & resources

Frequently asked questions

How to build a valid experimental design?

In Prostar, the differential analysis is devoted to the processing of hierarchical unpaired experimental designs. However, in former versions, this was not explicit enough, so that users with paired samples could used Prostar with wrong assumptions. To clear this out, we have changed the experimental design construction step so that its explicitly appears unpaired.

As a result, the samples must now be numbered as in the following example:

  • Condition 1: 1 - 2 - 3 - 4,
  • Condition 2: 5 - 6 - 7 - 8

As opposed to:

  • Condition 1: 1 - 2 - 3 - 4,
  • Condition 2: 1 - 2 - 3 - 4

Which, depending on the context, could suggest that the 8 samples comes only from 4 different biological subjects, and thus leading to paired tests - for instance, patients that are compared between Before (Condition 1) and After (Condition 2) some treatment.

However, one should note that even if the experimental design now looks different, this is just due to a numbering convention, and the statistical test is not impacted.

Why do the items of the contextual menus for plots remain ‘undefined’?

This happens if the version of the package ‘highcharter’ is less or equal to 0.5.0. To fix this issue, you should install the devel version of the package by typing the following command in a R console:

devtools::install_github('jbkunst/highcharter')

Why does my volcano plot look so aligned?

In very uncommun situations, one may obtain a bowl shape volcano plot such as depicted above. This is due to using Limma on a dataset for which it is not adapted: Briefly, the numerical values in the quantitative matrix appears to have a repetitive pattern that prevent Limma routines to compute the number of degrees of freedom of the Chi2 distribution on which the protein variances should be fitted. As a result, Limma returns a result directly proportional to the fold-change, and the p-values are none-informative. In such cases, which are fortunately extremely odd, we advise to replace Limma test by a classical t-test.

How to recover differential analysis results?

From Prostar 1.14, the differential analysis results are not exported anymore when using the “Export fo file” functionality, regardless the format (MSnset, excel or zipped CSV). This is due to the separate management of the “data mining” and “data processing” outputs. As a result, after performing the differential analysis, the results must be downloaded thanks to the devoted buttons (otherwise, they will be lost when closing the Prostar session). However, all the p-value computatations (from the “hypothesis testing” menu) can be exported and recovered from one session to another one.

Why isn’t it possible to adjust the logFC threshold on each differential analysis comparison?

Shortly, because from a statistical viewpoint, doing so roughly amount to FDR cheating. We have observed that numerous practitioners use the logFC threshold as a way to discard some proteins on the volcano plot, so that other proteins of interests appear more strikingly. In addition to be an uncontrolled and subjective way of sorting the proteins regardless of p-values, it has an important side effect on FDR computation: FDR computation requires a sufficiently large amount of proteins “below” the horizontal threshold on the volcano plot. However, all the proteins filtered out because of a too low logFC are not considered in FDR computation, so that tuning the logFC threshold to a too high value (so as to fine tune the protein selection) may lead to a spurious FDR. Finally, in case of more than two conditions (say A, B and C) , it would not make sense to defined differently (i.e. with a different logFC threshold) differentialy abundant proteins across various comparisons (e.g. when comparing AvsB, BvsC and CvsA). For all these reasons, we advise Prostar user to define once and for all the logFC threshold of each proteomics experiment (to a minimal value, such that below the threshold, a protein cannot be interesting froma biological viewpoint, because de FC cannot be properly exploited). More detailed explanations can be found in the following articles:

Why are the adjusted p-values hidden in the tabular outputs of the differential analysis?

Adjusted p-values (a. k. a. q-values) are often misunderstood. Because of their names, Prostar users often assume that an adjusted p-value is a protein-level piece of information, which quantitative value differs from the raw p-value, because of a mere “mathematical correction factor”, yet with essentially the same interpretation (that of a probabilistic quantification related to the DA status of the protein). However, this does not hold: an adjusted p-value is a list-related piece of information, which is exactly equivalent to an FDR: The adjusted p-value of the N-th protein with the smallest p-value equates the FDR when the list of putative DA proteins is cut to a length of N proteins. Consequently, a same protein with a same p-value, may have different adjusted p-values depending on the other p-values of the datasets (those which are smaller). This makes the adjusted p-value interpretation a bit touchy. In addition, displaying the adjusted p-value information in a tabular sheet were rows could be ordered in different ways (such as Excel spreadsheets for instance) is confusing: If the DA proteins are not sorted by increasing or decreasing p-values, their adjusted p-values cannot be related to the set of DA proteins they apply to. Thus, displaying them is both non-informative and error-prone. After having observed (or having received questions from) Prostar users who were misled by the adjusted p-value information, we have decided to remove it from the tabular outputs. An alternative would be to display the adjusted p-values, but with a different names, easier to interpret, such as “minimum FDR threshold at which the protein would be selected”. However, displaying this information promotes p-value hacking, as users may be tempted to spot their proteins of interest and tune the FDR accordingly. As we search for a better solution, any suggestion or comment is welcomed!

How the cells metadata tags are structured?

Prostar prior to 1.30.0

Peptide-level vocabulary

──Any
  │── 1.0 Quantitative Value
  │    │── 1.1 Identified
  │    │── 1.2 Recovered
  │
  │── 2.0 Missing value
  │    │── 2.1 Missing POV
  │    │── 2.2 Missing MEC
  │
  │── 3.0 Imputed value
  │    │── 3.1 Imputed POV
  │    │── 3.2 Imputed MEC

Protein-level vocabulary:

──Any
  │── 1.0 Quantitative Value
  │    │── 1.1 Identified
  │    │── 1.2 Recovered
  │
  │── 2.0 Missing value
  │    │── 2.1 Missing POV
  │    │── 2.2 Missing MEC
  │
  │── 3.0 Imputed value
  │    │── 3.1 Imputed POV
  │    │── 3.2 Imputed MEC
  │
  │── 4.0 Combined value

Prostar from 1.31.2

How the cells metadata tags are aggregated?

A set of rules define how the cells metadata tags are aggregated: Each example below represent a set of three peptides which are aggregated into one protein.

Note: Before an aggregation, a dataset does not contain:

  • x.0 tags: ‘Quantified’ (1.0), ‘Missing’ (2.0) nor ‘Imputed’ (3.0)
  • ‘Combined tags’ tag

These tags are specific to the aggregation process and appear only in aggregated proteins.

Rule 1: Aggregation of a unique type of quantified values (among 1.1, 1.2). If the type of all the peptides to agregate is either 1.1 or 1.2, then the resulting metadata tag is set to the corresponding tag

Rule 2: Aggregation of a unique type of missing values (among 2.1, 2.2). The aggregation of 2.1 peptides between each other gives a generic missing value (2.0)

Rule 3: Aggregation of a unique type of imputed values (among 3.3, 3.2). If the type of all the peptides to agregate is either 3.0, 3.1 or 3.2, then the resulting metadata tag is set to an ‘Imputed’ value (3.0).

Rule 4: Aggregation of a mix of quantified values (1.x). If the set of cell matadata tags to agregate is a mix of 1.x, then the final metadata is set to 1.0.

Rule 5: Aggregation of a mix of missing values (2.x). If the set of metacell to agregate is a mix of 2.x, then the final metadata is set to ‘Missing’ (2.0)

Rule 6: Aggregation of a mix of imputed values (3.x). If the set of cells metadata tags to agregate is a mix of 3.x, then the final metadata is set to ‘Imputed’ (3.0)

Rule 7: Aggregation of a mix of missing values (2.x) with any of quantitative and/or imputed values (1.x, 3.x). This case is not possible.

Rule 8: Agregation of a mix of quantitative values (1.x) and imputed values (3.x). If the set of metacell to agregate is a mix of 3.X and 3.0 and other (1.X), then the final metadata is set to ‘Combined tags’ (4.0).

Once the cell metadata tags have been aggregated, the final step of the process consist in analyzing missing values in order to identify ‘missing POV’ and ‘missing MEC’. Thus, at the end of the agregation process, a dataset may contains

Forum

Our community forum is hosted by the Bioconductor: https://support.bioconductor.org/t/prostar/


Community

Team presentation

Core team

Samuel Wieczorek

After a first career as IT support technician, Sam obtained an engineering degree (2004) at “Conservatoire National des Arts et Metiers”, followed by a MS degree in computer sciences and a PhD in machine learning (2009) at Grenoble-Alpes University. Since then, he has been working as a research engineer at EDyP-lab, where he has been developing and maintaining software tools for proteomics. Sam has been involved in Prostar project since its beginning. He is the code guru and supervises all the software aspects of the project, such as coding, packaging, deployment, debugging, graphical user interfaces, etc.

Thomas Burger

Tom is a CNRS senior scientist. He holds two MS degrees in computer sciences and applied mathematics (2004), a PhD in pattern recognition (2007) and a Habilitation thesis (2017), all from Grenoble Alpes University. Tom was an associate professor in machine learning with South Brittany University (2008-2011), before rushing back to his beloved mountains, with a CNRS position at EDyP-lab. He is the principal investigator of Prostar project. His expertise focuses on the statistical, methodological and algorithmic aspects of proteomics data analysis.

Contact us -

Occasional contributors

Beta-testing & co.

The entire EDyP proteomics platform (see www.edyp.fr): Prostar being permanently hosted by EDyP lab, the first users (the original ones, but also the testers) are naturally the lab members. They are all warmly acknowledged for their contributions.

Bug report

To report any issue with Prostar, it is best to use the devoted tab in Prostar software (click on Bug report in the Help menu), as it allows easy sharing of the session logs and data (essential to efficient debugging).

However, it is also possible to contact the development team by email (see team presentation).

Happiness report

If you are pleased with your Prostar experience, you can also send us a message (messages are not restricted to bug reports)!
:-)

Do not forget to cite Prostar in your publications!