than 50% of the fluctuations of the system. They can be used as
“reaction
coordinates”
to
generate
two-dimensional
(2D) probability distribution plots and identify putative conformational substates of different regions of the protein that need to be
predefined on the base of the structural knowledge on the system,
the question to be addressed or also a first visual exploration of the
MD trajectory. It is also useful to carry out all-atom PCA for a
specific region of interest upon fitting on the main-chain or CA of
the protein, to better appreciate the different states that these
regions can assume. The PCA results can also be mapped on the
3D structure of the protein, and they allow to identify a potential
coupling between different sites, as we observed for the conformation of the DNA-binding loop L1 (and its K120 residue especially)
and the distal loop S6-S7 [14].
3.5 Paths
of Communications
between Distal Sites
To further verify the coupling between the distal regions of interest,
i.e., the L1 and S6-S7 loops in our example, we can apply graph
theory to the MD structural ensembles. There are multiple solutions to achieve this goal available. Our group recently implemented a Python suite of tools that can handle different MD trajectory
formats, i.e., PyInteraph [68]. We also provided a plugin for Pymol
to handle PyInteraph output formats for graphical visualization of
the analyses [106]. As an alternative, the Wordom package could be
used which also allow integrating the PSN with information from
correlated motions [105]. PSN approaches such as the ones
provided by PyInteraph and Wordom can be used not only for
calculations of shortest communication pathways between two
residues but also to calculate other important properties of a network (such as hubs, connected components, cliques, etc.). We here
focus only on the path analyses since it is the suitable one for the
processes that this protocol aims at dissecting (i.e., communication
among distal sites).
We will here enter into the details of the PyInteraph approach,
which is currently used in our group. At first, we need to generate a
PSN based on contact between the centers of mass of residues side
chains using the pyinteraph tool of the package. A distance cutoff of
5 A ˚ has been recently shown as the best solution for this kind of
network in a benchmarking of different proteins simulated with
different force fields [75]. A Python tool, PyKnife, is also available to
identify the cutoffs and monitor the properties of the network with
a JackKnife resampling method [75]. The distance is calculated
between the center of mass of the residues side chains, expect for
glycines. To obtain the PSN, it is also recommended to retain those
edges that are only populated in the 20% of the simulation frames
[68, 73, 121] to remove noise in the final network. For each pair of
nodes of interest in the PSN graph (which could be, for example,
residues of L1 and residues of the S6-S7 loop), a variant of the
depth-first search algorithm is used in the current PyInteraph
234
Elena Papaleo
“reaction
coordinates”
to
generate
two-dimensional
(2D) probability distribution plots and identify putative conformational substates of different regions of the protein that need to be
predefined on the base of the structural knowledge on the system,
the question to be addressed or also a first visual exploration of the
MD trajectory. It is also useful to carry out all-atom PCA for a
specific region of interest upon fitting on the main-chain or CA of
the protein, to better appreciate the different states that these
regions can assume. The PCA results can also be mapped on the
3D structure of the protein, and they allow to identify a potential
coupling between different sites, as we observed for the conformation of the DNA-binding loop L1 (and its K120 residue especially)
and the distal loop S6-S7 [14].
3.5 Paths
of Communications
between Distal Sites
To further verify the coupling between the distal regions of interest,
i.e., the L1 and S6-S7 loops in our example, we can apply graph
theory to the MD structural ensembles. There are multiple solutions to achieve this goal available. Our group recently implemented a Python suite of tools that can handle different MD trajectory
formats, i.e., PyInteraph [68]. We also provided a plugin for Pymol
to handle PyInteraph output formats for graphical visualization of
the analyses [106]. As an alternative, the Wordom package could be
used which also allow integrating the PSN with information from
correlated motions [105]. PSN approaches such as the ones
provided by PyInteraph and Wordom can be used not only for
calculations of shortest communication pathways between two
residues but also to calculate other important properties of a network (such as hubs, connected components, cliques, etc.). We here
focus only on the path analyses since it is the suitable one for the
processes that this protocol aims at dissecting (i.e., communication
among distal sites).
We will here enter into the details of the PyInteraph approach,
which is currently used in our group. At first, we need to generate a
PSN based on contact between the centers of mass of residues side
chains using the pyinteraph tool of the package. A distance cutoff of
5 A ˚ has been recently shown as the best solution for this kind of
network in a benchmarking of different proteins simulated with
different force fields [75]. A Python tool, PyKnife, is also available to
identify the cutoffs and monitor the properties of the network with
a JackKnife resampling method [75]. The distance is calculated
between the center of mass of the residues side chains, expect for
glycines. To obtain the PSN, it is also recommended to retain those
edges that are only populated in the 20% of the simulation frames
[68, 73, 121] to remove noise in the final network. For each pair of
nodes of interest in the PSN graph (which could be, for example,
residues of L1 and residues of the S6-S7 loop), a variant of the
depth-first search algorithm is used in the current PyInteraph
234
Elena Papaleo