Figure legends
Figure 1: A hypothetical 16-state nonlinear QSP-style model. Each node
(circle) represents a system state (variable). Edges (solid black
arrows) represent mass (or molar) balance reactions and dashed lines
ending with bar are mass (or molar) action. (+) represents a positive
reaction, e.g. stimulation of a reaction, and (-) represents a negative
reaction, e.g., inhibition of a reaction.
Figure 2: A general framework for understanding different approaches to
model-order reduction. The original full-order model is shown as the
starting point. The model can then be reduced using either parametric
methods (shown as blue arrows) or nonparametric methods (shown in
orange). Parametric approaches produced a reduced-order model based on
mathematical techniques that stem from either reducing the number of
compartments (nodes) and/or reactions (edges). The nonparametric methods
approximate the full-order model input-output relationship (e.g. dose to
INR relationship) with a simpler empirical model, which might be a
black-box model input-output such as an artificial neural network or a
user defined empirical function.
Figure 3: An illustration of the concept of dimension reduction. (A)
3-dimensional surface showing a hypothetical relationship between
variables X,Y, and Z; (B) The same surface rotated clock-wise around the
Z-axis so that only X-Z perspective is visible. With a small margin of
error, Z can be approximated as a function of X, rather than as an
independent parameter, thereby reducing the order of the model.