Why Architecture Comes Before Kinetics
A delayed biomarker response may arise because a signal passes through several intermediate processes before it is measured. Before assigning time constants or fitting a response curve, the modeler must specify which processes are connected and where the measured signal enters and leaves the proposed system. That specification is the model’s architecture: its nodes, directed connections, inputs, and outputs.
Architecture is not merely a drawing convention. A serial chain says that the modeled signal passes through an ordered sequence of intermediate nodes. A branched pathway says that a node can influence more than one downstream route. These claims are structurally different, even if both diagrams can be adjusted to produce similar observed time courses. A schematic therefore records a mechanistic hypothesis; it does not, by itself, establish that the hypothesized causal connections are true.
A node need not correspond to a single molecule. It may represent a receptor-proximal signal, an aggregate intracellular process, a transducer, or a latent state that is not directly observed. Conversely, several measured quantities may be associated with one modeled node when different assays provide alternative or complementary observations of that state. The modeling choice should match the scientific question and the resolution supported by the data.
Use distinct labels for the external input, internal nodes, and observed outputs. For example, \(u(t)\) can denote an input signal at time \(t\), \(z_i(t)\) an internal node, and \(y_j(t)\) a measured output. An arrow between two modeled nodes represents a proposed direct connection in the chosen representation; it does not specify the connection’s sign, strength, timing, or mathematical form. Those are separate modeling decisions.
Representing a Cascade as a Directed Graph
A directed graph consists of a set of nodes and a set of oriented edges. A simple serial architecture can be written schematically as
Here the input enters the first modeled node, the internal nodes are ordered along one route, and the measured output is linked to the last node. This drawing states that the proposed route has no alternate path between the input and output. It does not require that the output be a direct reading of \(z_3(t)\); an assay may instead report a transformed, noisy, or otherwise indirect measurement of that state.
For clarity, a schematic should indicate whether a symbol denotes an internal state or an observation. An output can be drawn after an internal node to show where the assay is linked, but the output is not automatically another biological stage. Likewise, an external input is not usually counted as an internal transit stage. Stage counts are meaningful only after the modeler states what is being counted.
Worked Example: Counting Stages in a Serial Proposal
Suppose a hypothetical model proposes that an administered signal \(u(t)\) affects a receptor-proximal state \(z_1(t)\), then a second intracellular state \(z_2(t)\), then a downstream protein state \(z_3(t)\), which is assessed by assay \(y(t)\). The proposed schematic is
Step 1: Identify the input. The externally specified signal \(u(t)\) is the input. It is not one of the internal nodes in this representation.
Step 2: Count internal stages under the stated convention. The three internal states are \(z_1,z_2,z_3\), so this schematic has three modeled internal stages. The assay \(y(t)\) is the measured output, not a fourth biological stage.
Step 3: State the structural claim. There is one directed route from input to output, with three internal nodes in sequence. The diagram does not establish that each state is a distinct molecule, nor does it specify how quickly any state changes. Those claims would require additional biological and kinetic assumptions.
Serial, Branched, and Reconvergent Architectures
A serial chain has one ordered route through its internal nodes. In contrast, a branch occurs when a node has directed connections to two or more downstream nodes. The downstream routes may remain separate, or they may later reconverge at a shared node. A branched architecture can therefore contain parallel routes without implying that the routes are independent or that their effects simply add.
For a compact structural check, number the internal nodes and define an adjacency matrix \(A\) by setting \(A_{ij}=1\) when the schematic contains a directed edge from node \(i\) to node \(j\), and \(A_{ij}=0\) otherwise. This convention fixes the orientation: rows identify the source node and columns identify the destination node. The matrix records topology only; it does not contain rates, concentrations, or edge strengths.
Worked Example: Finding a Branch and Reconvergence
Consider a hypothetical cascade with internal nodes \(A,B,C,D\). The input \(u(t)\) enters \(A\); \(A\) connects to both \(B\) and \(C\); and both \(B\) and \(C\) connect to \(D\), where the measured output is linked. The internal-node edge list is \(A\to B\), \(A\to C\), \(B\to D\), and \(C\to D\).
Step 1: Count edges and identify branching. There are four internal edges. Node \(A\) has two outgoing edges, so the architecture branches at \(A\). Node \(D\) has two incoming edges, so the routes reconverge there.
Step 2: Write the adjacency matrix. In the order \(A,B,C,D\), the convention above gives
For instance, \(A_{13}=1\) records the edge \(A\to C\), while \(A_{31}=0\) indicates there is no edge \(C\to A\) in this proposal.
Step 3: Count routes from \(A\) to \(D\). The two directed paths are \(A\to B\to D\) and \(A\to C\to D\). Thus, there are two distinct two-edge paths from \(A\) to \(D\). This count is a property of the schematic, not a prediction that the output response will be twice as large or that both routes contribute equally.
The path-count result can be justified directly from matrix multiplication. The entry \((A^2)_{ij}\) equals \(\sum_k A_{ik}A_{kj}\). Each product is one when the edges \(i\to k\) and \(k\to j\) both exist, and zero otherwise. Therefore, \((A^2)_{ij}\) counts the two-edge routes from \(i\) to \(j\), including distinct routes through different intermediate nodes. In the example, \((A^2)_{AD}=A_{AB}A_{BD}+A_{AC}A_{CD}=1\cdot1+1\cdot1=2\). This is a topological calculation, not a kinetic derivation.
Worked Example: Two Outputs Do Not Always Mean a Branch
Suppose a hypothetical study reports two assay readouts, \(y_1(t)\) and \(y_2(t)\). The modeler considers two possible explanations. In the first, a single internal state \(Z\) is read by both assays. In the second, \(Z\) feeds two distinct downstream states \(B\) and \(C\), each measured separately:
Step 1: Classify the first representation. There is one internal node, \(Z\), and two observations linked to it. The two readouts alone do not establish a biological branch, because the model contains no two distinct downstream internal nodes.
Step 2: Classify the second representation. There are three internal nodes \(Z,B,C\). Since \(Z\) has two outgoing connections, this is a branched architecture. Each route has one downstream internal node after \(Z\), and the outputs are associated with different states.
Step 3: Explain what the data can establish. Observing two readouts is not sufficient to choose between these representations. The distinction depends on measurement properties, biological knowledge, and whether the time-course data provide evidence for separate downstream behavior. Without such evidence, presenting the branch as established would overstate what the observations show.
Assumptions, Diagnostics, and Common Failure Modes
A schematic is interpretable only when its scope and conventions are stated. A useful architecture specification makes at least four choices explicit:
- Node meaning: what biological or functional quantity each node represents, and whether it is observed or latent.
- Edge meaning: which directed influences are included, and whether arrows denote hypothesized direct influence rather than mere correlation.
- Input placement: where the externally specified signal enters the system and whether other external drivers are omitted or represented.
- Output mapping: which node or combination of nodes is linked to each assay, including whether an assay is an indirect measurement.
Architecture also rests on simplifications. A diagram may omit feedback, unmeasured routes, or interactions with other systems. It may treat a complex process as one node, or represent a distributed process as several nodes. These are modeling decisions, not discoveries implied by the drawing. If a proposed pathway includes feedback, its graph is not a simple acyclic chain; the schematic should show the feedback edge rather than force the system into a serial ordering.
Diagnostics for structural trouble begin with the diagram and the data together. Check whether every internal node lies on a route relevant to an input or measured output. A disconnected node, an unexplained output, or an arrow with no scientific rationale may signal an unnecessarily complicated or incomplete representation. Compare the predicted ordering of observed changes with the proposed route: persistent timing patterns inconsistent with a single serial route may motivate consideration of alternative connections, but timing alone does not prove a branch. Also ask whether the measured outputs can distinguish two candidate architectures. If both structures imply the same observable quantities under the available measurements, the data do not identify which structure is correct.
A frequent mistake is to infer a branch from the fact that two biomarkers change after the same input. The flawed reasoning is: “Both outputs respond, so the input must split into two biological pathways.” The correction is that shared response can arise from a common internal state measured twice, from distinct downstream routes, or from other unmodeled causes. A branch is a claim about distinct modeled connections; it needs supporting evidence or should be presented explicitly as a hypothesis.
Choosing an Architecture Without Overclaiming
The main alternative to a detailed cascade is a more aggregated representation, such as one effective intermediate node between input and output. Aggregation can be appropriate when intermediate states are not measured and the scientific goal is to describe the overall delay rather than assign it to specific components. A more detailed serial or branched diagram is useful when biological knowledge or multiple measurements support the intermediate structure. Greater detail is not automatically better: every added node or edge introduces a structural claim that may be weakly informed by the data.
These alternatives differ in interpretation, not merely in diagram complexity. A serial chain proposes one route through ordered stages. A branched graph proposes multiple downstream routes from a shared node. A shared-state measurement model proposes multiple observations of one internal process rather than multiple biological routes. The appropriate choice depends on the question, the available assays, and prior biological evidence. The architecture alone does not determine a unique set of dynamic equations, parameter values, or response predictions.
A careful modeling report should therefore state the proposed nodes, edge directions, input location, and observation links before presenting a kinetic model. It should distinguish established biological knowledge from structural assumptions made for analysis. This keeps a schematic useful: it communicates exactly what is being modeled while leaving the mathematical description of signal evolution to the subsequent specification of the model.
Check Your Understanding
Answer each question from the stated schematic conventions. Short solutions follow the questions.
- A schematic is \(u\to P\to Q\to y\), with \(P\) and \(Q\) internal states. How many internal stages are shown, and is the architecture serial or branched?
- In a directed graph, \(A\to B\), \(A\to C\), \(B\to D\), and \(C\to D\). Identify the branch point, the reconvergence node, and the number of paths from \(A\) to \(D\).
- Two assays report different values, but both are modeled as observations of one latent state \(Z\). Does this alone establish a branched biological pathway? Why or why not?
- With nodes ordered \(A,B,C\), a schematic has edges \(A\to B\) and \(A\to C\). Write its adjacency matrix using rows as sources and columns as destinations.
- What does an arrow in a proposed cascade establish, and what does it not establish?
Solutions. (1) There are two internal stages, \(P\) and \(Q\); the route is serial. (2) \(A\) is the branch point, \(D\) is the reconvergence node, and there are two paths: \(A\to B\to D\) and \(A\to C\to D\). (3) No. Two measurements of one modeled state are not two distinct downstream biological routes. (4) The matrix is \(\begin{pmatrix}0&1&1\\0&0&0\\0&0&0\end{pmatrix}\). (5) It records a directed influence proposed by the model; by itself it does not establish a causal mechanism, the sign or strength of influence, or its kinetics.