Biology is notoriously messy. At every scale—from single-molecule interactions to whole-organism metabolism—living systems are defined by high-dimensional, non-linear, and stochastic processes. To make sense of this complexity, computational biologists build mathematical models. However, a model cannot be as complex as the reality it describes; if it were, it would be as intractable as the system itself. As Lin et al. note, computational models are inherently simplifications bound by specific assumptions, meaning each can only address a subset of biological problems.
This creates a fundamental tension in systems biology: how much biological complexity can we omit before a model loses its predictive power? Understanding why models simplify, how they abstract reality, and where these abstractions break down is essential for building reliable predictive frameworks.
The Necessity of Abstraction: Tractability and Identifiability
Why do we simplify? The primary driver is mathematical and computational tractability. If a model includes every known receptor, phosphorylation site, and secondary messenger, it becomes impossible to parameterize.
Witt et al. demonstrate this trade-off by showing that models prioritizing parameter identifiability must omit secondary biological processes like protein hyperphosphorylation or de novo synthesis. Without these simplifications, the model's parameters cannot be uniquely determined from experimental data, rendering the model practically useless for quantitative prediction.
Furthermore, when detailed molecular data is scarce, simplified models serve as crucial conceptual foundations. Agarwal et al. argue that even first-generation simplifications—such as omitting specific signaling substances or biochemical regulatory effects in skin tanning—can offer valuable insights and guide future experimental designs. Simplification is not merely a limitation; it is a deliberate strategy to uncover core mechanisms that would otherwise be obscured by biological noise.
Common Strategies for Omitting Complexity
Computational models typically reduce biological complexity through three main mechanisms:
1. Lumping Components and Steps
Instead of modeling every intermediate state, multiple biological entities or multi-step pathways are treated as a single variable. For example, Stumpf simplifies the E. coli stress response by treating two distinct proteins, PspB and PspC, as a single complex. Similarly, Cepeda-Humerez et al. reduce a multi-step biological proofreading mechanism to a single energy-consuming step, while Giampieri et al. treat entire protein complexes and miRNA clusters as single elements to maintain model tractability.
2. Truncating Regulatory Networks
Models often focus exclusively on a handful of "master regulators" while ignoring auxiliary pathways. Holmes et al. point out that cell fate specification models frequently utilize simplified gene regulatory networks that focus on a limited set of master regulators, omitting secondary transcription factors and auxiliary feedback loops. This reductionist view is also highlighted by Lio et al., who note that incorporating only a subset of molecular interactions can overlook critical signaling pathways responsible for tissue degradation and recovery. Similarly, Heath et al. observe that many potentially significant regulatory reactions are routinely omitted from signaling pathway models to maintain simplicity.
3. Ignoring Spatial and Cellular Heterogeneity
Many models assume a well-mixed environment or uniform cell behavior, ignoring the spatial architecture of tissues. In mathematical oncology, Lipková et al. and Weidner et al. observe that mechanistic tumor growth models regularly rely on simplified equations that overlook vascular network development, necrosis, and intratumoral heterogeneity.
This omission extends to tissue-level modeling as well; Du et al. explain that multiscale models often focus on primary cell types while omitting rare populations, which limits their ability to capture the full range of physiological signaling interactions within the tissue.
The Risks of Over-Simplification
While abstraction makes models solvable, omitting too much regulatory or structural complexity introduces significant risks, particularly when models are applied to clinical or evolutionary predictions.
- Loss of Predictive Precision: In drug delivery and radiotherapy, over-simplification can lead to suboptimal treatment plans. Hwang et al. argue that continuum models of drug-eluting stents simplify cellular internalization and proteolysis, necessitating more detailed modeling to predict local drug distribution accurately. For glioblastoma, Unkelbach et al. note that simplified tumor growth models fail to fully capture the intricate dynamics of in vivo proliferation and migration, directly impacting spatial dose redistribution in radiotherapy planning.
- Misrepresenting System Dynamics: Omitting ecological or environmental factors can distort how we understand disease progression and resistance. Maltas et al. show that failing to explicitly incorporate the immune system or spatial cell arrangements limits our understanding of how drug resistance emerges. Similarly, Zaika et al. demonstrate that omitting distinct biological pathways, such as the serrated polyp pathway in colorectal cancer, restricts the precision of clinical screening recommendations.
- Overlooking High-Dimensional Behavior: Mehrjou emphasizes that simplified dynamical systems often fail to capture the high-dimensional gene regulatory networks and intricate spatial structures inherent in real tumors. When these dimensions are stripped away, models can miss critical phase transitions, non-linear feedbacks, or emergent properties like metabolic self-regulation (Chintaluri) and redundant resistance mechanisms (Danish et al.).
Finding the Right Level of Complexity
How do researchers determine the optimal boundary of a model? The consensus across recent literature is that models must evolve alongside our experimental capacity to measure the omitted variables. Jüngel suggests that mathematical frameworks for vascular network formation must be systematically extended to incorporate cell-cell interactions and extracellular matrix remodeling to match biological reality.
Interestingly, biology itself might offer a solution to the problem of high dimensionality. Kaneko demonstrates through spin-glass modeling that complex biological systems naturally evolve to exhibit "dimension reduction," where phenotypic responses to environmental and genetic changes become simple and predictable due to evolved robustness to noise. By identifying these naturally reduced dimensions, computational biologists can build models that are both mathematically simple and profoundly accurate.
Key Foundational Work
Inverse metabolic engineering: A strategy for directed genetic engineering of useful phenotypes (1996) — contributed 1 concept elements to this direction.
Impact of transport and drug properties on the local pharmacology of drug-eluting stents (2003) — contributed 1 concept elements to this direction.