The new Bioconductor release

The new version (1.4) of the BioNAR package is released. In this version we add the number of important features:

Edge weights

From version 1.3.5 the edge weights can be taken into account in clustering and centrality measure calculations by weight parameter or graph edge attribute. The value of weight parameter must be a positive numeric vector, NULL or NA. If it is NULL and the input graph has a ‘weight’ edge attribute, then that attribute will be used. If the weight parameter is NULL and no such attribute is present, then the edges will have unit weights. Set the weight parameter to NA if the graph has a weight edge attribute, but you don’t want to use it. A larger edge weight means a stronger connection for this function. The weights value is ignored for the spectral clustering algorithm.

There is one important nuance that have to be taken into account: different algorithms treat weights differently. For example, in the betweenness calculations edge weight is treated as edge length: the higher the weight the longer the distance between nodes and the lower the probability that this edge will be part of the shortest path. At the same time, in the Page Rank “an edge with a larger weight is more likely to be selected by the surfer”, which infer the opposite meaning.

Taking into account that all implemented clustering algorithms treat edge weight as an interaction strength measure, i.e. in the same way as Page Rank algorithm we use this meaning as the default. So for all versions of distance-based centrality measure calculations such as betweenness (BET, dBET) and shortest path statistics (mnSP and sdSP) we calculate the distance as

for consistency.

DYNAMO perturbation pattern calculations

In 2018 Santolini and Barabasi publish a paper in which they show that topology of interaction graph can up to some extent reproduce sensitivity patterns of underlying dynamical system. They call their algorithm DYNAMO (DYNamics-Agnostic Network MOdels). For us that algorithm is interesting as it allow negative edge weights to represent inhibition interactions within the network. We rework their original code from [[https://github.com/msantolini/dynamo/]] to use sparse matrices, which is necessary to work with large graphs.

Decoupling from synaptome.db and synaptome.data packages

In the process of refactoring the code for execution in a parallel manner in HPC environment we realise that dependency from synaptome.db and because of it from synaptome.data makes initialisation of the computing environment extremely time consuming. Taking into account that the only reason for these dependencies is the construction of PPI networks via querying synaptome database, we decided that the dependency should be reversed: network creation should be created within synaptome.db package with the help of BioNAR functionality. So from version 1.4 BioNAR do not use any code from the synaptome.db.