secondary structure. We performed the calculation of salt
bridges considering the distances between positively and negatively charged groups, as by default (see Note 4 for more
details). In this case, we decided to keep all the edges with a
persistence value higher than zero (see Note 8).
2. Identify the persistence significance threshold (P crit ). In order
to remove the most transient interactions, we used the filter_graph tool to perform an estimation of the threshold of significance for the interaction persistence in the graph (P crit ). P crit is
calculated by filtering the original graph several times for
increasing persistence threshold values, which is accomplished
by removing edges having a weight lower than a given cut-off.
For each filtered graph, the size of the biggest connected
component is then calculated. The plot resulting from the
procedure, as the one from the graph of hydrophobic contacts
shown in Fig. 1a, generally exhibits an abrupt decrease, often
with a sigmoid-like shape for globular proteins, with a central
point that represents a threshold with a good balance between
having a too interconnected and a too sparse network. Based
on this analysis, we set a persistence threshold of 20, which was
also compatible and used for the salt bridges and hydrogen
bond graphs (not shown). This threshold indicates the minimum percentage of frames in which an interaction should be
present over the whole ensemble to be considered.
3. Filter each graph according to the identified persistence threshold. We used the filter_graph program to perform graph filtering according to the selected P crit threshold of 20, removing all
the edges with weight lower than this value. We specified the
graph to be filtered with the option -d and the persistence
threshold using the -t option. Figure 1b–d shows the filtered
salt bridges, hydrogen bonds, and hydrophobic cluster networks, respectively.
4. Visualization and analysis of the calculated networks. We plotted and visualized each interaction type on the reference structure, using the xPyder plug-in for PyMOL. To do so, we loaded
the topology PDB file in PyMOL, opened the xPyder plug-in,
and loaded an adjacency matrix file. We then generated the
corresponding graph in the Graph analysis tab and visualized
hubs, defined as residues involved in three or more distinct
interactions.
It should be noted that network properties, such as hubs and
connected components, can be analyzed using either xPyder or
the graph_analysis tool in PyInteraph. See section 3.5 for an example of the latter.
Salt bridges can have local and long-range effects in proteins,
especially in solvent-exposed and disordered regions that have a
Interaction Networks with PyInteraph
161
bridges considering the distances between positively and negatively charged groups, as by default (see Note 4 for more
details). In this case, we decided to keep all the edges with a
persistence value higher than zero (see Note 8).
2. Identify the persistence significance threshold (P crit ). In order
to remove the most transient interactions, we used the filter_graph tool to perform an estimation of the threshold of significance for the interaction persistence in the graph (P crit ). P crit is
calculated by filtering the original graph several times for
increasing persistence threshold values, which is accomplished
by removing edges having a weight lower than a given cut-off.
For each filtered graph, the size of the biggest connected
component is then calculated. The plot resulting from the
procedure, as the one from the graph of hydrophobic contacts
shown in Fig. 1a, generally exhibits an abrupt decrease, often
with a sigmoid-like shape for globular proteins, with a central
point that represents a threshold with a good balance between
having a too interconnected and a too sparse network. Based
on this analysis, we set a persistence threshold of 20, which was
also compatible and used for the salt bridges and hydrogen
bond graphs (not shown). This threshold indicates the minimum percentage of frames in which an interaction should be
present over the whole ensemble to be considered.
3. Filter each graph according to the identified persistence threshold. We used the filter_graph program to perform graph filtering according to the selected P crit threshold of 20, removing all
the edges with weight lower than this value. We specified the
graph to be filtered with the option -d and the persistence
threshold using the -t option. Figure 1b–d shows the filtered
salt bridges, hydrogen bonds, and hydrophobic cluster networks, respectively.
4. Visualization and analysis of the calculated networks. We plotted and visualized each interaction type on the reference structure, using the xPyder plug-in for PyMOL. To do so, we loaded
the topology PDB file in PyMOL, opened the xPyder plug-in,
and loaded an adjacency matrix file. We then generated the
corresponding graph in the Graph analysis tab and visualized
hubs, defined as residues involved in three or more distinct
interactions.
It should be noted that network properties, such as hubs and
connected components, can be analyzed using either xPyder or
the graph_analysis tool in PyInteraph. See section 3.5 for an example of the latter.
Salt bridges can have local and long-range effects in proteins,
especially in solvent-exposed and disordered regions that have a
Interaction Networks with PyInteraph
161