Fig 1.
Overview of subnetwork inference task.
The first panel shows a background network composed of molecular entities and known interactions among them. A subset of the entities are designated as source nodes (black), and a subset as target nodes (blue). White nodes show elements that are not considered source or target nodes. The second panel displays an inferred subnetwork which includes only a subset of the entities and interactions in the background network. Elements of the background network that are not included in the inferred subnetwork are greyed out.
Fig 2.
General representation of a background network augmented with information from scientific literature.
The background network is composed of source nodes (black), target nodes (blue), and interactions (dashed and solid lines). Elements outlined in red represent information extracted from the scientific literature. In this case, the literature has been used to identify another source node (A), an additional target (V), and two additional interactions.
Fig 3.
General representation of a view generated from a subnetwork.
The first panel shows an inferred subnetwork. The second panel shows a view (yellow background) which delineates a set of nodes and interactions that are related to a concept of interest. The inferred subnetwork and view are composed of source nodes (black), target nodes (blue), and interactions (dashed and solid lines). Elements outlined in red represent entitities and interactions used as the basis of the view (i.e., those most closely linked to the concept).
Fig 4.
Overview of the HIV host-virus subnetwork inference task.
The first panel displays a background network composed of source RNAi hits (black), interface nodes (green), and interactions (dashed and solid lines). Each interface is connected to a target viral component (blue). We can optionally augment the set of sources by using genes identified by GADGET as being relevant to HIV replication (red outline). The second panel displays an inferred subnetwork. Elements of the background network that are not included in the inferred subnetwork are shown in light grey.
Table 1.
Objective functions and constraints for the host-virus integer program.
The left column describes each objective function and constraint. The right column provides the mathematical formulation for each.
Fig 5.
Overview of the NR2F1 subnetwork inference task.
The first panel displays a background network composed of the source NR2F1 (black), target DE genes (blue), and interactions (dashed and solid lines). Transcription factors are identified as bright magenta nodes. The background network is augmented by a set of Literome interactions (purple arrows). The second panel displays the inferred subnetwork. Orange nodes show elements that are not considered source, transcription factors, or target nodes; we refer to these as intermediate nodes in the inferred subnetwork.
Table 2.
Objective functions and constraints for the NR2F1 integer program.
The left column describes each objective function and constraint. The right column provides the mathematical formulation for each.
Fig 6.
Precision-recall curves for HIV replication subnetworks.
The blue line represents results from the consensus network without GADGET genes augmenting the sources. The red line represents results from the consensus network generated with GADGET augmentation. The horizontal light green line in the figure represents the prevalence of known hits and interfaces in the background network.
Fig 7.
Precision-recall curves for HIV replication subnetworks with and without literature extracted genes and interactions.
The blue line represents results from the consensus network without GADGET or Literome augmentation. The red line represents results from the consensus network generated with GADGET genes augmenting the sources. The light blue line represents results from the consensus network generated with Literome interactions augmenting the network. The orange line represents results from the consensus network generated with both GADGET genes and Literome interactions augmenting the sources. The horizontal light green line in the figure represents the prevalence of known hits and interfaces in the background network.
Fig 8.
The number of subnetwork genes that are within one interaction away from a set of known essential basal cell genes from Marcotte et al. [45].
The magenta line represents genes included the Literome-augmented subnetworks, and the green line represents genes included in the baseline subnetworks which did not contain literature extracted information.
Fig 9.
The number of subnetwork genes that are in (A) the top 10% in RNA-Seq abundance, and (B) the top 25% in RNA-Seq abundance.
The magenta line represents genes included in the Literome-augmented subnetwork, and the green line represents genes included in the baseline subnetwork which did not contain literature extracted information.
Fig 10.
Graphical representation of the inferred HIV subnetwork and the elements of it that are selected for the membrane scission view.
Light grey nodes represent nodes within the subnetwork that are not part of the view. Black nodes represent RNAi hits (source nodes), green nodes represent host interface genes, dark grey nodes represent elements that are both RNAi hits and interfaces, and blue nodes represent viral components (target nodes). Nodes with red borders are the GADGET hits used to anchor the view.
Fig 11.
Graphical representation of a view onto the HIV subnetwork for the membrane scission concept.
Black nodes represent RNAi hits (source nodes), green nodes represent host interface genes, grey nodes represent elements that are both RNAi hits and interfaces, and blue nodes represent viral components (target nodes). Nodes with red borders are the GADGET hits used to anchor the view.
Fig 12.
Graphical representation of a view onto the HIV subnetwork for the concept intrinsically disordered proteins.
Black nodes represent RNAi hits (source nodes), green nodes represent host interface genes, grey nodes represent elements that are both RNAi hits and interfaces, and blue nodes represent viral components (target nodes). Nodes with red borders are the GADGET hits used to anchor the view.