Please paste your MS/MS spectrum into the field below:

[Optional] Please paste the mass differences into the field below:

Messages from the server:

Here is the list of annotated features (masses and mass differences):

Here is the ranked list of suggested substructures:


Welcome to MESSAR, the substructure recommendation tool!

This is an automated approach for substructure recommendation from MS/MS spectra of (unknown) small metabolites.

Our goal is not to identify the structure of the entire molecule, but to propose potential substructures to experts, helping them better characterize the unknown e.g. identifying functional classes, grouping similar spectra and supporting NMR-based de novo identification etc.

Substructure recommendations are computed based on 8378 associations between spectral features and structural features. We call these associations “rules”. These rules are discovered from GNPS spectral library ( containing mass spectra of about 4000 known metabolites using association rule mining.


MESSAR is developed for processing single MS/MS spectrum in positive ion mode.

A) Start a run

Figure 1

1. MS/MS spectrum is compulsory for running the web tool. It must be in either format: i) one column of m/z value ii) two columns m/z - intensity, separated by tab or space.

2. Mass differences (losses) used for substructure prediction can be specified (values in one column). It is useful when users are only interested in certain mass differences in the spectrum. If not specified, all mass differences in the spectrum will be used to compute substructures.

3. Precursor m/z is a recommended input (if known by users). It can be important because when MS/MS spectra do not contain the precursor ion, mass differences between precursor and fragments i.e. neutral losses can be overlooked.

4. Relative intensity threshold is used to filter low-intensity peaks: peaks below the intensity threshold (percentage of the highest peak) will not be used for substructure recommendation

5. N most intense peaks are used for substructure prediction. Please put 0 if all peaks are considered and spectra filtering will be only based on relative intensity threshold.

6. Error (ppm) allowed when annotating spectral features (experimental masses) with our rules. We note that our rules are associations between exact masses and substructures.

7. Please click on “Submit” to start substructure recommendation. User can start a complete new run or stop at anytime during the analysis by clicking on “clear” button.

8. Once the analysis is complete, a message will be displayed. Users are invited to check the results in tab panels B and C if substructures are recommended. There will be an error message if the input format is not valid or no substructure is found. In the latter case, the problem might be solved by decreasing the relative intensity threshold or by increasing tolerance window.

B) Annotated features

The tab panel displays all MESSAR rules that matched with MS/MS features of input spectrum. It can also be seen as substructure annotations of every MS/MS feature

Table 2.

1. Summary of number of rules and total number of substructure recommended

2. A table of all matched rules:

  • TYPE: type of spectral feature in the body of the rule. For example, “[mass, mass_diff]” means that the rule predicts the substructure based on the co-presence of a fragment and a loss in the input spectrum.

  • FEATURE: m/z values of corresponding fragments or losses

  • SUBSTRUCTURE: predicted substructure

  • SCORE[SENSITIVITY]: sensitivity measures the statistical significance of matched rules. The higher the score is, the more meaningful is the rule for substructure prediction.

C) Sub-structure suggestions

The tab panel displays all annotated and scored substructures after aggregating rules by one of the three algorithms.

Table 3.

1. The method used to aggregate rules to substructures. User can choose between Fast/Naive/Exhausive. Algorithms Fast and Exhausive both cluster rules according to similarity of substructures, and retrieve the maximum common substructure (MCS). Algorithm Fast only use top 20 rules (see previous tab-panel) to perform aggregation and is therefore faster. The algorithm Naive simply combines rules that suggest IDENTICAL substructures. For all three algorithms, the score of each substructure is the sum of sensitivity of all responsible rules. We recommend users to try out all three algorithms to discover the most reliable and meaningful substructures.

2. A ranked list of substructure recommendations

3. Score distribution of substructures. The arrow indicates the score of substructure selected by user in the table.

4. Input MS/MS spectrum. The red peaks and arrows indicate fragments and losses that are responsible for the substructure selected by user.


Developped by Youzhong LIU (

Department of Mathematics and Computer Science, University of Antwerp, Antwerp, Belgium

Biomedical Informatics Network Antwerpen (biomina), University of Antwerp, Antwerp, Belgium

You can find source code of this site here.

Powered by R Shiny.