Inference of the dynamic protein-potein interaction network in the context of aging
Gene expression analyses can identify highly expressed (active, dysregulated)
genes at the given condition (age, time point, biological process, patient).
However, these analyses typically study genes (their protein products) in
isolation, ignoring their interconnectivities. But proteins carry out cellular
function by interacting with each other. Thus, analyses of protein-protein
interaction networks (PINs), which model these interactions, are promising.
However, these analyses ignore any condition-specific information, because
the current PINs span different conditions. Clearly, integration of
condition-specific gene expression data with PIN data to construct
condition-specific PINs is promising.
Indeed, recent studies have done this. They map the expression levels of active
genes to their corresponding proteins in the PIN, propagate the expression
information throughout the network, assign weights to nodes (proteins) and
edges (protein-protein interactions) in the network based on the propagated
expression information, and identify highly weighted network regions as
condition-specific PINs. However, these network propagation (NP) methods
results in static condition-specific PINs.
In contrast, cellular processes
are dynamic. This includes aging, which is important to study because incidence
of serious diseases increases with age. Hence, studying aging, a dynamic
process, via static network analysis is suboptimal. Can we instead construct
dynamic, age-specific PINs, to study the evolution of network structure and
thus cellular functioning with age?
The existing approach for dynamic PIN
inference maps genes that are active at the given age according to the gene
expression data onto the static PIN while keeping all interactions among the
genes (i.e., it extracts the induced subgraph on the active genes) to get
the PIN specific to that age. Then, it repeats this for each age. However,
the induced approach has drawbacks. First, not all interactions between the
active genes are necessarily equally “important”, and hence, we want some
mechanism to identify and keep only the most important interactions. Second,
we want to consider both genes that are active as well as genes that are not
necessarily active but that critically connect the active genes. NP can help
So, we generalize the existing notion of static NP to its
dynamic counterpart and construct NP-based (rather than induced) dynamic,
age-specific PINs. Then, we predict genes whose PIN positions significantly
change with age as aging-related. We show that our dynamic NP
improves the accuracy of the aging-related predictions compared to the
existing static NP and the existing dynamic induced approach.
Thus, dynamic NP will likely impact future research of dynamic biological
networks, beyond “just” the study of aging.