Why¶

Full Correlation Analysis (FCA) aids resolve conformational substates and identify actual transition pathways, usually much more successfully so than Principal Components Analysis (PCA).

Algorithm¶

FCA of a trajectory follows the spirit of PCA in finding important modes of motion through an orthonormal coordinate transformation and subsequent sorting of basis vectors.
While the projection of the trajectory on the subspace spanned by any two PCA basis vectors is linearly uncorrelated, they usually share mutual information (think of a circle).
The flavour of FCA to implement here, minimizes mutual information in the total system as defined by the information entropy H, I=\sum H[x_i]-H[{x}].
It starts from an intial guess of Eigenvectors from a PCA, then iteratively minimizes Mutual Information by 2d-planes rotations of pairs of basis vectors.

Implementation¶

The algorithm described here has been implmented for gromacs3 by Oliver Lange and then ported to gromacs4.6 by Christian Blau.
It should fit perfectly the trajectoryanlysis framework

References¶

Lange, Oliver F., and Helmut Grubmüller. "Full correlation analysis of conformational protein dynamics." Proteins: Structure, Function, and Bioinformatics 70.4 (2008): 1294-1312.

[[http://www.mpibpc.mpg.de/9619680/g_fca_ver_1_3.tgz]]
[[http://www.mpibpc.mpg.de/9768579/g_fca_ver_2_0.tgz]]