Modern biology is currently facing a "snapshot" problem. While technologies like single-cell RNA sequencing and mass spectrometry provide high-resolution views of cellular states, they typically capture only a single moment in time. As Marinka Zitnik and colleagues argue in Machine learning for integrating data in biology and medicine: Principles, practice, and opportunities, most multi-modal integration frameworks are static, failing to account for how biological systems evolve or change their state in response to treatment.
To move beyond these static descriptions, researchers are developing a new suite of computational tools designed to infer the underlying temporal dynamics of biological networks. The goal is to treat the cell not as a fixed list of parts, but as a dynamic system governed by evolving control logic.
The Evolving Wiring Diagram
In a static gene regulatory network, we might see that Gene A influences Gene B. However, in reality, this "wiring" is often transient. A regulatory relationship might only exist during a specific phase of the cell cycle or in response to a specific environmental trigger.
To capture this, Eric P. Xing and Le Song introduced Time-Varying Dynamic Bayesian Networks (TV-DBN). By combining kernel smoothing with sparse regression, this framework allows the directed edges of a network to change over time. This approach reveals functional groups and interactions that are typically averaged out or completely obscured by static models, enabling researchers to track the "re-wiring" of the genome as a disease progresses.
Cells as Dynamical Systems
If we view the cell as a machine, we can apply the mathematics of control theory to understand its behavior. Momiao Xiong and colleagues pioneered this in Differential Dynamic Properties of Scleroderma Fibroblasts in Response to Perturbation of Environmental Stimuli, treating gene regulatory networks as systems that can be modeled using the same differential equations used to stabilize aircraft.
This perspective is particularly powerful for understanding drug resistance. For example, Fuhai Li and Jielin Xu developed a model that combines static expression with Hidden Markov Models (HMMs) to identify "loops" in signaling pathways. They found that effective treatments successfully collapse these feedback loops, while resistant cancers use them to bypass drug-induced blockades. By modeling the system dynamically, they could predict how networks reorganize in response to a drug—something a static snapshot cannot do.
Tracking Cellular Conversations
Biological tissues are not just collections of cells; they are networks of constant communication. Current spatial transcriptomics methods often produce static maps of these interactions, but as Cameron Y. Park and Shouvik Mani point out, these "conversations" wax and wane.
Their framework, DIISCO, uses Gaussian Process Regression Networks to infer how intercellular interactions evolve over time from irregularly sampled clinical data. This allows researchers to see not just who is talking to whom, but how the intensity and nature (activating vs. inhibitory) of those signals change during disease progression. Similarly, tools like CrossTalkeR by James S. Nagai apply graph theory to single-cell data to quantify how these communication networks shift between healthy and diseased states.
Synthesizing Pathways from Time-Series Data
One of the greatest challenges in this field is the "dark matter" of signaling—the vast majority of protein interactions that are not captured in canonical databases. Ali Sinan Köksal addressed this by developing the Temporal Pathway Synthesizer (TPS).
TPS uses formal logic and constraint solving to reconstruct signaling pathways directly from temporal phosphoproteomic data. By integrating time-series mass spectrometry with known interactions, it can map how signals propagate across the entire proteome in real-time. This moves the field away from universal, fixed pathway maps toward context-dependent, dynamic models that reflect the specific environment of a living cell.
Why Intuition Matters
The shift toward dynamic modeling is more than a mathematical upgrade; it is a conceptual shift in how we define a "disease state." If a disease is a trajectory rather than a static condition, then a successful therapy is one that redirects that trajectory. As argued in Whither systems medicine? by Rolf Apweiler and others, understanding these underlying temporal dynamics is the only way to design combinatorial therapies that can truly outmaneuver the adaptive mechanisms of complex diseases like cancer.