Welcome to the fpocket web server

Welcome to the fpocket [1,2] web server, dedicated to centralize information on how to run fpocket-based online server tools.

Three servers are proposed and accessible from this page (see on the right):

  • Fpocket: perform simple pocket detection
  • MDpocket: track pockets in molecular dynamics
  • Hpocket: view conserved pockets withing homologous proteins

There are actually two version of each server: a default and an advanced version (using Mobyle portal). Both make use of exactly the same protocol, and will return you exactly the same results. Only, the mobyle portal has several advantages over a simple standalone server, like the possibility to bookmark jobs, register as user to store jobs on the server (parameters included), and use pipelining features to transfer F/H/MDpocket output to any other compatible service available in mobyle.

Fpocket suite

fpocket is a very fast, open source protein pocket (cavity) detection algorithm based on Voronoi tessellation. It was developed in the C programming language and is currently available as command line driven program, and from now as a web server too.

Since its first release, the new package mdpocket has been developped. This package has been designed to track cavities on different, aligned protein structures, and thus can be used to:

  • track protein cavities during molecular dynamics trajectories (original goal)
  • search for conserved pockets in protein families

Currently available as a web server only, mdpocket will be released in the fpocket official distribution as soon as the corresponding scientific paper will be published.

If you use fpocket for publication, please cite:

Schmidtke P, Le Guilloux V, Maupetit J, Tuffery P, fpocket: online tools for protein ensemble pocket detection and tracking , Nucleic Acids Res. 2010 May 16.
Le Guilloux V, Schmidtke P and Tuffery P, Fpocket: An open source platform for ligand pocket detection , BMC Bioinformatics, 2009, 10:168

Quick overview of web servers

Predict potential protein pockets!

Based on one or several input PDB files, the Fpocket web server will return you a set of potential pockets under various output formats, including:

  • Input protein output as PDB file including alpha spheres (which represent pockets) as dummy carbon atoms
  • PDB and PQR files defining each predicted pockets
  • Executable visualisation scripts for VMD and Pymol

In addition, several snapshots are automatically generated for visualisation purposes, and results can also be analysed directely on the servuer using Jmol and OpenAstex applets, as shown in the following two example snapshots.


Track pockets during molecular dynamics


The MDPocket online server allows you to use fpocket capabilities on molecular dynamics trajectories to detect highly conserved void zones during the simulation. All you need is your trajectory under the form of multi, superposed PDB snapshots. Similarly to fpocket, various files will be returned:

  • MDpocket grid (.dx file), representing alpha spheres density (see methods)
  • The input structure output as PDB file, with the bfactor value of each atom representing alpha sphere density around this atom. High B-factors represent atoms lining highly conserved pockets. Several pictures are given as output to represent this information, as shown in the snapshot on the right (red color = hotspots!)

The interactive part of the results visualisation is performed using either Jmol or OpenAstex applets. The protein is shown along with the mdpocket output grid represented as meshed isosurface. You can then play with the isovalue: highest isovalue will display surfaces corresponding to highly conserved pockets. Lowest isovalues will display all places where alpha spheres have been found during molecular dynamics. Similar operation can be done using B-factor coloring on each atoms. See the mddpocket documentation to learn more about that.

Importantly, MDpocket also allows you to monitor a specific pocket for each of your snapshots. See the input and output sections of the online documentation.


View conserved pockets in homologous structures


Based on a query protein, the Hpocket server allows you to retrieve homologous proteins having 3D structures available (using BLAST), and track conserved pockets within all homologues retrieved using the MDpocket methodology.

The visualisation of results is pretty much the same as that described previously for MDpocket (protein alignements snapshots example shown on the right). Hpocket will also retrieve the following information:

  • Blast results, as a report txt file
  • PDB Hits sequence alignments, in fasta format.
  • An archive containing superimposed PDB homologues



Fpocket is among the fastest pocket prediction methods. Running fpocket on a single protein will take less than one second to 3-4 seconds, depending on the protein size. Although the server isn't dedicated to such purpose, Fpocket is thus well suited for large scale studies (pocketome). As MD and H pocket are build on top of Fpocket, they are also very fast.


Fpocket has been evaluated and compared to other recent geometry-based methods (see the BCM paper) on external diverse protein-ligands complexes. The following figure on the right shows that the fpocket performance is comparable, sometimes better than other geometric methods on both apo and holo proteins.

Density Density


To our knowledge, MDpocket is the first free and open source tool allowing to track cavities on a set of PDB files. The server applies the methodology to homology structures as well as MD trajectories, providing usefull tools to analyse pockets conservation, dynamics and hot spot.


Fpocket is very well documented. In addition to the BMC paper, a full (PDF) document gives several basic examples on how to run various fpocket programs, and explain in details the various parameters. It also contains several code snippet examples explaining how to include your own scoring function, to add a new pocket descriptor... The package also includes an API documentation.


Each fpocket algorithm parameter can be changed via. command line parameters. Each command line parameters is explained in the documentation, or in the unix manuals created at fpocket installation (type *man fpocket* in a shell).

Free & opensource

fpocket is totally free and opensource (GPL licence).

<Back to top>


Transientness is implicitely tracked by MDpocket

Transient pockets are not explicitely tracked. They are defined by low isovalue clusters of grid points, but you need to use the second round of MDpocket to gain insight into the real dynamic of a given pocket.

Depends on Fpocket clustering procedure

MD and Hpocket are based on Fpocket. They are both sensible to parameters used by fpocket (this is also an advantage as changing parameters allows to track different type of pockets)

No energy criteria used

Both Fpocket and H/MDpocket are exclusively geometry based. In our opinion, geometric methods have the advantage of tracking protein parts where small molecules could bind from a steric, geometric point of view. This is an advantage as we make no assumption about what types of interaction could occure. However, we believe that energetic methods could improve pockets delimitation by removing highly unvaforable zones.

High structural differences

This is a theorical limitation, not seen yet in practice.. MDpocket methodology depends the quality of the superposition between different structures. If homologous proteins are structurally too different, or if the protein motions are too important during molecular dynamics, results might not be satisfying.

<Back to top>



fpocket is a geometric based pocket detection algorithm which is based on the concept of alpha spheres, introduced by Liang and Edelsbrunner [3] and also used by Chemical Computing Group in the SiteFinder software [4]. It makes use of the QHULL library [5] to perform Voronoi tesselation to gather alpha spheres.

Briefly, an alpha sphere is a sphere that contacts four atoms on its boundary and contains no internal atom. Alpha sphere radii reflect the local curvature defined by the four atoms: 4 atoms in a plane would correspond to an alpha sphere of infinite radius, and conversely, 4 atoms packed at the apex of a tetrahedron would lead to a value of radius close to that of the Van der Waals radius. For a protein, very small spheres are located within the protein, large spheres at the exterior, and clefts and cavities correspond to spheres of intermediate radii.

The fpocket algorithm can be divided in the following steps:

  • Calculate alpha spheres on the whole protein
  • Filter out alpha spheres with radius m < R < M, with m = 3.0 and M = 6.0 in the current implementation
  • Cluster alpha sphere using a multiple linkage clustering algorithm
  • Merge nearby pockets
  • Remove small pockets
  • Score and rank pockets

Pocket scoring

Once all pockets have been clustered, they are ranked using a simple scoring function making use of several dpocket descriptors. This scoring function have been calibrated using a training set of 307 known complexes (see the paper for more information).


MDpocket is the first official extension of the fpocket 1.0 release, and allows you to track protein pockets during molecular dynamics.

Briefly, based on a set of aligned PDB structures, mdpocket will run fpocket on each of these snapshots and collect all resulting alpha spheres. Once done, a discrete grid will be created (1A resolution by default) and superposed to all alpha spheres. The number of alpha spheres around each grid point will be counted, resulting in a pocket density grid, similar to an electron density map.

For each grid point, mdpocket counts the number of alpha spheres located around a sphere of 8A (default parameters). Thus, the more a cavity is conserved during the MD, the higher this value will be for a given grid point (alpha sphere density).


The hpocket concept is basically the application of MDpocket to homologous structures. As MDpocket takes standard PDB files as input, we can apply the MDpocket methodology to any set of PDB files, with the necessity of them to be structurally superposed, and that you know what you are doing (aligning completely unrelated proteins does not make much sense...)

Based on this observation, we created a web service, Hpocket, dedicated to make easier the application of MDpocket to homologous structures. Note that the alignement step is even more important to obtain interpretable results.

<Back to top>


To get support, you can either send an email to the fpocket support mailing list, or send email directly to the fpocket team, using emails given in the BMC bioinformatics paper. You may also want to subscribe to the newsletter to be informed of news and releases.

<Back to top>


[1] Le Guilloux V, Schmidtke P and Tuffery P, Fpocket: An open source platform for ligand pocket detection , BMC Bioinformatics, 2009, 10:168

[2] Schmidtke P, Le Guilloux V, Maupetit J, Tuffery P, fpocket: online tools for protein ensemble pocket detection and tracking , Nucleic Acids Res, 2010, 2010 May 16

[4] The Chemical Computing Group [ http://www.chemcomp.com/ ]

[5] Barber CB, Dobkin DP and Huhdanpaa HT, The Quickhull algorithm for convex hulls , ACM Trans. on Mathematical Software, 1996, 22(4):469-483
<Back to top>


fpocket has been developed by Peter Schmidtke and Vincent Le Guilloux during a MSc project proposed and supervised by Dr. Pierre Tuffery. MDpocket has been developed by Peter Schmidtke and Hpocket preprocessing by Julien Maupetit. Online documentation has been written by Vincent Le Guilloux. All online services have been developed by Julien Maupetit.

<Back to top>