DTU Health Tech

Department of Health Technology

GibbsCluster - 2.0

Simultaneous alignment and clustering of peptide data



GibbsCluster is a server for unsupervised alignment and clustering of peptide sequences. The program takes as input a list of peptide sequences and attempts to cluster them into meaningful groups, using the algorithm described in this paper.
Visit the links on the grey bar below to read instructions and guidelines, see output formats, or download the code.

Update (Nov 2016): Implements deletions and insertions in the sequence alignment.

For very large data sets, you are encouraged to download a stand-alone version of the program, with full functionality and no parameter limitations.

Submission


DATA SUBMISSION

Paste peptides in the box:

or submit a file directly from your local disk:



To load some SAMPLE DATA click here:

More sample data: Sample 1 - Sample 2 - Fibroblast (Bassani, MCP 2016)

SUBMIT job



Optionally set some of the parameters before starting your job, or load one of the recommended configurations:

MHC class I ligands of same length:
MHC class I ligands of length 8-13:
MHC class II ligands:

BASIC options

Hover the mouse cursor over the symbol for a short description of the options

Job name

Number of clusters

Motif length


ADVANCED options

Make clustering moves at each iteration

Max deletion length

Max insertion length

Number of seeds for initial conditions

Penalty factor for inter-cluster similarity (λ)

Weigth on small clusters (σ)

Use trash cluster to remove outliers


VERY ADVANCED options

Number of iterations per sequence per temperature step

Initial Monte Carlo temperature

Number of temperature steps

Interval between Indel moves

Interval between Single Peptide moves

Interval between Phase Shift moves

Background amino acid frequencies

Preference for hydrophobic AAs at P1

Sequence weighting type



SUBMIT job



Confidentiality:
The sequences are kept confidential and will be deleted after processing.


CITATIONS

  • GibbsCluster: unsupervised clustering and alignment of peptide sequences
    Andreatta M, Alvarez B, Nielsen M
    Nucleic Acids Research (2017) doi: 10.1093/nar/gkx248
    Full text: PUBMED
  • Simultaneous alignment and clustering of peptide data using a Gibbs sampling approach
    Andreatta M, Lund O, Nielsen M
    Bioinformatics (2012) doi: 10.1093/bioinformatics/bts621
    Full text: PUBMED

Instructions & Guidelines



GibbsCluster takes as input a list of peptide sequences and attempts to cluster them into meaningful groups.
Sequence alignment and clustering are performed simultaneously by sampling the space of possible solutions using a Gibbs sampling strategy. Each cluster is represented by a position-specific scoring matrix (PSSM), and the algorithm aims at maximizing the information content of individual matrices while minimizing the overlap between distinct clusters. The server returns a detailed report on the optimal clustering solutions, including plots of the optimal number of clusters and graphical representations of the identified sequence motifs as sequence logos. For details on the algorithm refer to this paper

This page introduces the data formats, the parameters available to customize the analysis and some guidelines for the use of version 2.0. Users are welcome to contact the authors for any questions.

1. Specify INPUT sequences

Paste a set of peptides (up to 50 amino acids in length), one sequence per line into the upper left window, or upload a file from your local disk.

The input file can be a plain list of peptides (Sample 1) or an annotated list (Sample 2). The annotation is carried over to the results, and may be useful for correlating a known classification with the clustering produced by the method. All input sequences must be in one-letter amino acid code. The allowed alphabet (not case sensitive) is as follows:

A C D E F G H I K L M N P Q R S T V W Y

2. Set OPTIONS to customize your analysis

The options are divided in three levels: Basic, Advanced and Very advanced. The only essential parameters to be set are in the Basic options; if you are using the server for the first time you may leave the advanced options unchanged.

A brief explanation of each option can be visualized by hovering the mouse over the symbol next to each option in the submission form.

BASIC options

Job name:
This prefix is pre-pended to all files generated by the current run. If left empty, a system-generated number will be assigned as prefix.

Number of clusters:
You may provide a specific number of clusters (e.g. 3), or an interval of partitions (e.g. 1-8). In the second case, the method will suggest the optimal number of cluster it found in the data, given the parameter configuration of the job. Maximum number of clusters: 15.

Motif length:
The algorithm will attempt to align all sequences to common windows of N amino acids, and construct its PSSMs on these alignments. Specify with this option the length of the alignment window. Minimum motif length: 2

ADVANCED options

Make clustering moves at each iteration:
By default, simple shift moves are performed at each iteration, indel moves every 10 iterations, single peptide moves every 20 iterations, phase shift moves every 100 iterations.
You can alter this behavior by ticking this option; simple shift and phase shift moves become disabled, and single peptide moves are made at each iteration. This set-up is recommended for "nearly-aligned" data, where clustering and indels should be sampled more regularly than extensions at the termini. That is the case, for example, of sets of MHC class I ligands of different length, which would in most cases require central indels to model peptide bulging of long ligand.

Max deletion length:
The maximum length of consecutive deletions in a peptide sequence.

Max insertion length:
The maximum length of consecutive insertions in a peptide sequence.

Number of seeds for initial conditions:
Gibbs sampling is a heuristic rather than a rigorous optimization procedure. Therefore, it cannot guarantee that the most optimal solution is always reached from any starting configuration. A common procedure to boost performance is to repeat the sampling from a number of initial random configurations and select the solution that appears to be optimal in terms of the fitness function that governs the system. Specify with this parameter the number of initial configurations used to initialize the system.

Penalty factor for inter-cluster similarity (λ):
This parameter modulates how similar the clusters are allowed to be. If you believe your data contains multiple specificities with well-defined motifs, λ can be relatively high; on the other hand, if your aim is to detect subtle differences in mostly homogenous data, the parameter λ should be set to a lower value.

Weigth on small clusters (σ):
This parameter can be used to specify how small clusters are allowed to be. With low values of σ the method will tend to produce small specialized clusters, while larger σ will return larger and more general clusters.

Use trash cluster to remove outliers:
The trash cluster is used to collect the peptides that appear not to match any of the motifs being identified. The behaviour of the trash-cluster is identical to any of the other clusters, with the difference that the sequences in the trash cluster do not contribute to the overall score of the system.

Threshold for discarding to trash:
This parameter specifies a baseline on the peptide scores, below which peptides are tossed into the trash cluster. If you believe your data contains some degree of noise, you may experiment with increasing this value and observe how many sequences become filtered out by the trash cluster.

VERY ADVANCED options

Number of iterations per sequence per temperature step:
This parameter ("I") specificies how long your clustering schedule should be. Note that total number of iterations is the results of "I" multiplied by the number of sequences times the number of temperature steps, and it will increase linearly the execution time.

Initial Monte Carlo temperature:
The temperature is a scalar, lowered by discreet steps as the iterations progress. The temperature influences the probability of accepting or rejecting the moves of the algorith. In the initial iterations (high temperature) the program is free to explore the landscape of solutions, and as the system cools off only moves that increase the energy will be accepted.

Number of temperature steps:
The number of steps in the cooling schedule (starting from the initial temperature specified above).

Interval between Indel moves:
Specifies how often to attempt introducing insertions and deletions (see glossary).

Interval between Single peptide moves:
Specifies how often to attempt moving a sequence between clusters (see glossary).

Interval between Phase shift moves:
Specifies how often to attempt shifting the alignment window of a single cluster.

Background amino acid frequencies:
Construction of PSSMs relies on calculating the frequency of a given residue at a given position, compared to the expected background frequency of that amino acid. You may use a flat background model identical for all amino acids (Flat), a pre-calculated distribution reflecting the relative frequency of each residue in naturally occurring proteins (Pre-calculated Uniprot), or determine the background model directly from the dataset you submitted (From data).

Preference for hydrophobic AAs at P1:
In the special case of MHC class II data, we have previously found helpful to guide the alignment by expressing a preference for hydrophobic residues at the P1 of the alignment.

Sequence weighting type:
Data redundancy may affect the quality of the clustering. You may use an explicit clustering of the sequences in a given group (Clustering), or use a faster heuristic that calculates the degree of variability at each column in the alignment (Heuristic, recommended); you may also disable sequence weighting for downweighting of redundant sequences (None).

3. SUBMIT the job

Click on the "Submit query" button. The status of your job (either 'queued' or 'running') will be displayed and constantly updated until it terminates and the server output appears in the browser window.

At any time during the wait you may enter your e-mail address and simply leave the window. Your job will continue; you will be notified by e-mail when it has terminated. The e-mail message will contain the URL under which the results are stored; they will remain on the server for 24 hours for you to collect them.

GLOSSARY

  • PSSM: Position-specific scoring matrix. A matrix of size L x A, where L is the length of the alignment window and A is the length of the alphabet; it stores the weight (or "preference") or any given amino acid at each position of the alignment.
  • Indels: Insertions and deletions. An insertion adds one or more gaps in the alignment, a deletion removes one or more amino acids of a given sequence.
  • Simple shift move: A move affecting the alignment core of a single peptide. The algorithm attempts to shift the alignment core of a peptide to the left or right, and accept/reject the move with a probability that depends on the temperature of the system.
  • Single peptide move: A move that attempts to change the cluster of a single peptide sequence.
  • Indel move: A move that introduces or remove an insertion or a deletion in a peptide sequence. In the GibbsCluster, indels can have any length up to the values specified as parameters, but there can only be one single indel stretch per peptide (i.e. a single sequence cannot contain indels at multiple positions)
  • Phase shift move: A move affecting the alignment core of all peptide sequences in a cluster. In simple terms, the algorithm will attempt to shift en bloc the alignment window of a given cluster.

Output format



DESCRIPTION

An example of output is found below. The output is composed of the following sections:
  1. Run identifiers and settings
  2. The parameters specificied by the user are reported here, together with the number of sequences loaded as input.

  3. Barplot of KLD vs number of clusters
  4. For each initial number of clusters, the information content of the alignments is shown as a barplot. The relative size of each block within a bar is proportional to the size of a given cluster. In the example below, the run with 3 initial clusters produced one empty cluster, therefore only 2 boxes are depicted in the third column of the barplot.

  5. Sequence logos of the optimal solution
  6. The sequence motifs identified by the Gibbs Clustering are shown to the right of the barplot. This is the optimal solution over the range of initial numbers of clusters. Hovering the cursor over the logos shows the KLD of each cluster in the solution.

  7. Complete results for all initial number of clusters
  8. If a range of initial numbers of clusters was specified (1 to 3 in the example below), the results for each case are listed in succession. The optimal number of cluster shown above is just a suggestion and depends on the specificied parameters, so the user is encouraged to inspect all solutions with different numbers of clusters.

    It is possible to customize the logos further by clicking on the LOGO button. This transfers the data to the Seq2Logo server, which allows plotting several different kinds of sequence logo.

    Inspect the complete Clustering Report and the formatted Clustering Solution for a tabular version of the results. Format of the Clustering Solution files:
    • Gn: Cluster number
    • Num: Sequential number of the sequence in the cluster
    • Sequence: Complete peptide sequence
    • Core: Portion of the peptide in the alignment window
    • of: Offset value, i.e. the starting position of the alignment core
    • IP: Position of the insertion, if any
    • IL: Length of the insertion, if any
    • DP: Position of the deletion, if any
    • DL: Length of the deletion, if any
    • Annotation: Text annotation to the peptide, if provided at submission
    • Self: Score of the peptide to its own cluster
    • bgG: Identifier of the nearest cluster (-1 means no nearest cluster)
    • bgScore: Score of the peptide to the nearest cluster
    • cScore: corrected score (Sself - λ x Snearest)


    Remember that results are only stored on the CBS server for about 24 hours. Save your results to disk by clicking on the DOWNLOAD link at the bottom of the results page.



EXAMPLE OUTPUT



   

GibbsCluster Server - Results

Technical University of Denmark





Version: 2.0
Run ID: 27911
Run name: gibbs_27911
Platform: Linux x86_64


Read 200 unique sequences from file

Settings:
No shift moves, cluster moves at every iteration
Number of clusters: 1 - 3
Motif length: 9
Initial MC temperture: 0.8
Number of temperature steps: 20
Number of iterations x Sequence x Tstep: 100
Max insertion length: 1
Max deletion length: 5
Interval between Indel moves: 10
Number of initial seeds: 3
Penalty lambda: 0.8
Weight on small clusters: 10
Sequence weighting type: 0
Background model: Uniprot pre-calculated
Use trash cluster to remove outliers: 1
Threshold for trash cluster: 0

KLD vs. Number of clusters with λ = 0.8
Identified 2 sequence motifs

View the barplot in full size


RESULTS for 1 CLUSTERS

Final Average KLD: 9.466122

 Group   Size   KLD   Seq2Logo   Matrix 
 1   197   9.466 
  Mat_1.1  
 Outliers   3         

Raw Clustering Report
Formatted Clustering Solution
Clustered Alignment Cores

RESULTS for 2 CLUSTERS

Final Average KLD: 10.917174

 Group   Size   KLD   Seq2Logo   Matrix 
 1   100   9.372 
  Mat_1.2  
 2   100   12.462 
  Mat_2.2  
 Outliers   0         

Raw Clustering Report
Formatted Clustering Solution
Clustered Alignment Cores

RESULTS for 3 CLUSTERS

Final Average KLD: 10.907671

 Group   Size   KLD   Seq2Logo   Matrix 
 1   101   12.348 
  Mat_1.3  
 2   99   9.438 
  Mat_2.3  
 3   0   0.000 
 Outliers   0         

Raw Clustering Report
Formatted Clustering Solution
Clustered Alignment Cores



See the Activity log for this job

Article abstract


GibbsCluster: unsupervised clustering and alignment of peptide sequences

Massimo Andreatta, Bruno Alvarez, Morten Nielsen

Nucleic Acids Research, 2017 Apr 12. doi: 10.1093/nar/gkx248


Receptor interactions with short linear peptide fragments (ligands) are at the base of many biological signaling processes. Conserved and information-rich amino acid patterns, commonly called sequence motifs, shape and regulate these interactions. Because of the properties of a receptor-ligand system or of the assay used to interrogate it, experimental data often contain multiple sequence motifs. GibbsCluster is a powerful tool for unsupervised motif discovery because it can simultaneously cluster and align peptide data. The GibbsCluster 2.0 presented here is an improved version incorporating insertion and deletions accounting for variations in motif length in the peptide input. In basic terms, the program takes as input a set of peptide sequences and clusters them into meaningful groups. It returns the optimal number of clusters it identified, together with the sequence alignment and sequence motif characterizing each cluster. Several parameters are available to customize cluster analysis, including adjustable penalties for small clusters and overlapping groups, and a trash cluster to remove outliers. As an example application, we used the server to deconvolute multiple specificities in large-scale peptidome data generated by mass spectrometry. The server is available at http://www.cbs.dtu.dk/services/GibbsCluster-2.0.

Full text

Software Downloads




GETTING HELP

If you need help regarding technical issues (e.g. errors or missing results) contact Technical Support. Please include the name of the service and version (e.g. NetPhos-4.0) and the options you have selected. If the error occurs after the job has started running, please include the JOB ID (the long code that you see while the job is running).

If you have scientific questions (e.g. how the method works or how to interpret results), contact Correspondence.

Correspondence: Technical Support: