gmx cluster can cluster structures using several different methods. Distances between structures can be determined from a trajectory or read from an .xpm matrix file with the -dm option. RMS deviation after fitting or RMS deviation of atom-pair distances can be used to define the distance between structures.
single linkage: add a structure to a cluster when its distance to any element of the cluster is less than cutoff.
Jarvis Patrick: add a structure to a cluster when this structure and a structure in the cluster have each other as neighbors and they have a least P neighbors in common. The neighbors of a structure are the M closest structures or all structures within cutoff.
Monte Carlo: reorder the RMSD matrix using Monte Carlo such that the order of the frames is using the smallest possible increments. With this it is possible to make a smooth animation going from one structure to another with the largest possible (e.g.) RMSD between them, however the intermediate steps should be as small as possible. Applications could be to visualize a potential of mean force ensemble of simulations or a pulling simulation. Obviously the user has to prepare the trajectory well (e.g. by not superimposing frames). The final result can be inspect visually by looking at the matrix .xpm file, which should vary smoothly from bottom to top.
diagonalization: diagonalize the RMSD matrix.
gromos: use algorithm as described in Daura et al. (Angew. Chem. Int. Ed.1999, 38, pp 236-240). Count number of neighbors using cut-off, take structure with largest number of neighbors with all its neighbors as cluster and eliminate it from the pool of clusters. Repeat for remaining structures in pool.
When the clustering algorithm assigns each structure to exactly one cluster (single linkage, Jarvis Patrick and gromos) and a trajectory file is supplied, the structure with the smallest average distance to the others or the average structure or all structures for each cluster will be written to a trajectory file. When writing all structures, separate numbered files are made for each cluster.
Two output files are always written: -o writes the RMSD values in the upper left half of the matrix and a graphical depiction of the clusters in the lower right half When -minstruct = 1 the graphical depiction is black when two structures are in the same cluster. When -minstruct > 1 different colors will be used for each cluster. -g writes information on the options used and a detailed list of all clusters and their members.
Additionally, a number of optional output files can be written: -dist writes the RMSD distribution. -ev writes the eigenvectors of the RMSD matrix diagonalization. -sz writes the cluster sizes. -tr writes a matrix of the number transitions between cluster pairs. -ntr writes the total number of transitions to or from each cluster. -clid writes the cluster number as a function of time. -cl writes average (with option -av) or central structure of each cluster or writes numbered files with cluster members for a selected set of clusters (with option -wcl, depends on -nst and -rmsmin). The center of a cluster is the structure with the smallest average RMSD from all other structures of the cluster.