🧬 biology

What a description must preserve to compare system responses

This paper synthesizes how specific descriptions must preserve particular relations to accurately predict system responses across changes, illustrating through diverse mathematical and biological examples that maintaining invariants or repeated measurements does not guarantee adequacy without accounting for history, internal states, and nonlinearity.

Original authors: Elkin Antonio Navarro Quiroz

Published 2026-10-12

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Elkin Antonio Navarro Quiroz

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

In the study of living systems, scientists often face a choice: describe a system in all its intricate detail, or summarize it with a few broad numbers. Imagine trying to understand a forest. You could count every single leaf, track the health of every root, and measure the soil moisture at every inch. Or, you could simply measure the total weight of the forest. Both descriptions contain information, but they serve different purposes. The total weight might tell you how much carbon the forest stores, but it cannot tell you if a specific disease is spreading through the canopy or if the trees are struggling to find water. This tension between detail and simplicity is at the heart of how we model biology. Researchers have long known that sometimes, a simple summary is enough to predict how a system will behave. But they have also known that this is not always true. The critical question is not just whether a summary works, but exactly what information must be kept, and what can be safely thrown away, to ensure the prediction remains accurate when the system changes.

A new analysis by Elkin Antonio Navarro Quiroz at Universidad Simón Bolívar tackles this question by examining the specific conditions under which a simplified description fails. The work does not propose a new universal law or a single algorithm to solve every biological problem. Instead, it acts as a rigorous map of the boundaries where our summaries break down. The author uses carefully constructed mathematical models to show that two systems can look identical when described by a simple total, yet behave completely differently when challenged. For instance, the paper demonstrates that knowing the average amount of a substance in a group of cells is often insufficient if the cells respond to that substance in a non-linear way. If one cell has a heavy load and another has a light load, the total might be the same as a group where every cell has a medium load, but the biological outcome for the first group could be very different. The paper proves that without knowing how the burden is distributed among the individuals, a prediction based on the average alone can be dangerously wrong.

The research goes deeper, showing that even when we know the average for every type of cell, we can still be misled if we do not know how those types are paired together. The author illustrates this with a scenario involving two attributes, such as the number of opportunities a cell receives and its efficiency in using them. If you only record the total opportunities and the total efficiency for a population, you miss the crucial detail of which cells got the opportunities. A population where the most efficient cells receive the most opportunities will produce a very different result than a population where the least efficient cells get the most opportunities, even though the totals are identical. The paper establishes that the "pairing" of these attributes is a piece of information that must be preserved to make an accurate prediction. Without it, the description is incomplete, regardless of how precise the totals are.

To make these abstract failures concrete, the author builds a specific model of a system with two compartments, like two rooms exchanging material. In this model, material flows in, moves between the rooms, and leaks out at different rates. The study shows that two different setups can have the exact same amount of material in each room and the exact same total flow leaving the system, yet they will react differently to a sudden change in input. One setup might hold onto the material, while the other might flush it out quickly. The difference lies in the hidden history of how the material moved and the specific rates of exchange that are invisible if you only look at the final amounts. The paper demonstrates that if you try to predict the future behavior of these systems based only on their current total amounts, you will fail. The missing piece of information is the internal state or the history of the system, which acts as a memory that influences the future.

The analysis also explores the role of measurement itself. It shows that the way a scientist chooses to measure a system can create an illusion of stability or change. In the two-compartment model, a sensor might be set up to measure a specific combination of the amounts in the two rooms. Depending on how the sensor is tuned, it might show a dramatic difference between two systems, or it might show no difference at all, even though the systems are behaving differently internally. The paper argues that a measurement that looks stable does not necessarily mean the system is stable; it might just mean the measurement is blind to the changes happening inside. This distinction is vital for understanding biological signals, where a flat line on a monitor might hide a complex and dangerous internal shift.

Furthermore, the author addresses the concept of symmetry, often thought to imply that a system is balanced and predictable. The paper shows that a system can have symmetrical rules but still behave asymmetrically if it started from an uneven state. Conversely, a system can look symmetrical in its totals while hiding a complex, asymmetric exchange of materials. The work clarifies that observing a symmetry in the output does not guarantee that the underlying mechanism is simple or that the system is robust. It serves as a warning against assuming that because a pattern looks regular, the processes driving it are simple.

The paper concludes by proposing a way forward for biological research, specifically in the context of repeated cellular events like the uptake of dead cells. The author suggests that to truly understand these processes, researchers must move beyond simple summaries of the current state. They need to design experiments that test whether knowing the history of an event adds value to predicting the next one. The study outlines a framework for testing this: compare a model that only looks at the present with one that includes the past, and see if the history actually improves the prediction on new, independent data. The author is careful to note that this is a proposal for how to test these ideas, not a claim that history is always necessary. In some cases, the current state might be enough; in others, the past is essential. The key is to know which case you are in before you make a prediction.

Ultimately, this work is a guide to the limits of simplification. It does not tell scientists to stop using summaries, but rather to understand exactly what they are losing when they do. It establishes that a description is only as good as the specific question it is trying to answer. If the question is about the total amount of material, a simple sum might suffice. If the question is about how the system will respond to a change, that sum might be dangerously inadequate. The paper provides the logical tools to distinguish between these situations, ensuring that when scientists simplify a complex biological system, they do not accidentally discard the very information needed to understand it. The findings are derived from mathematical models and simulations, offering a clear set of rules for when a description is sufficient and when it is not, leaving the application to real-world biological data as the next step for researchers to take.

Technical Summary: What a description must preserve to compare system responses

Problem Statement
The paper addresses a fundamental challenge in systems biology and modeling: determining what specific relations a system description must retain to accurately predict a system's response as conditions change. The author argues that the utility of a description (a summary of state, such as a total, a marginal distribution, or a vector of attributes) is not intrinsic but is strictly relative to the specific response being predicted (a quantity, trajectory, or probability distribution) under defined inputs, time horizons, and contexts. A central issue is that descriptions often fail to preserve responses because they omit critical distinctions—such as internal distributions, pairings of attributes, or historical states—even when they preserve aggregate totals or invariants. The paper seeks to separate the tasks of choosing a description, preserving a response, and explaining the underlying mechanisms, demonstrating that response preservation alone does not identify a unique mechanism or guarantee predictive adequacy.

Methodology
The paper employs a critical synthesis of established mathematical and theoretical frameworks, including information theory, computational mechanics, causal inference, and dynamical systems theory. The methodology relies on constructing specific counterexamples and analytical comparisons to test the sufficiency of various descriptions. Key methodological components include:

Mathematical Definitions of Sufficiency: The paper defines an "exactly sufficient description" $g(\xi)$ for a response $R$ such that $g(\xi_A) = g(\xi_B) \implies R(\xi_A) = R(\xi_B)$ . It distinguishes this from statistical sufficiency and mechanistic identification.
Counterexample Construction: The author constructs specific scenarios to demonstrate insufficiency:
- Nonlinear Averaging: Using independent units with nonlinear response functions to show that equal total burdens do not yield equal mean responses (Jensen's inequality).
- Marginal vs. Joint Distributions: Using a bilinear model to show that equal marginal distributions of attributes (e.g., opportunities and efficiencies) can yield different aggregate outputs if the pairing (covariance) is altered.
- Information Synergy: Using independent bits to demonstrate that joint information can be necessary for a target even when individual sources provide zero information.
- Compartmental Dynamics: A two-compartment model with exchange and unequal loss rates is used to analyze how internal redistribution affects retention and outflow, and how measurement weights can conceal or create signal contrasts.
- Symmetry and Invariance: A rotation example using the Cayley map is used to show that an exact invariant (norm preservation) can coexist with accumulating phase errors, failing to predict position-dependent targets.
Error Bounds: The paper derives lower bounds on prediction error (Equation 2) when two systems share a description but yield different responses, establishing that a description fails a required precision threshold even before learning algorithms are applied.

Key Contributions
The paper does not propose a new universal reduction algorithm or a new biological law. Instead, its contribution is a synthesis of "target-specific failures and repairs," clarifying the conditional nature of model reduction and description adequacy.

Distinction Between Preservation and Explanation: The work clarifies that preserving a response (sufficiency) does not identify the causal mechanism. A description can be sufficient for a specific output without revealing the internal processes or history that generated it.
Specific Sources of Insufficiency: The paper identifies four primary ways descriptions fail to preserve responses:
- Heterogeneity: Nonlinear individual responses mean that mean inputs do not predict mean outputs.
- Pairing: Discarding the association between attributes (e.g., covariance) can alter aggregate outputs even if marginal distributions remain fixed.
- History Dependence: Eliminating internal states (like a compartment's internal contrast) can introduce exact history dependence, requiring initial-state terms or higher-order derivatives to close the dynamics.
- Measurement Artifacts: Sensor weighting can create or cancel signal contrasts independent of the actual biological retention or outflow.
Calibration and Repair: The paper proposes that sufficiency can be achieved through calibration on admitted perturbation directions. For linear systems, a calibrated response to specific redistribution vectors can predict stationary responses without identifying every internal rate, provided the domain of perturbations is restricted.
Operational Criteria for Adequacy: The paper outlines that a description's adequacy depends on the declared horizon, the response metric, the tolerance for error, and the biological unit of analysis (e.g., distinguishing between independent preparations and cells within a preparation).

Results
The analytical results demonstrate that:

Equal totals are insufficient: In nonlinear systems, equal total inputs do not guarantee equal responses; the distribution of the burden matters.
Marginals are insufficient: In systems with bilinear interactions, the covariance between attributes determines the output, meaning separate knowledge of marginals is inadequate.
Invariants are not guarantees: A description can preserve a mathematical invariant (like a norm or a ratio) while failing to preserve a dynamic target (like position or phase), as shown in the rotation example.
Stationary equivalence is weak: Systems with identical stationary totals and outflows can have different internal states and different transient responses to perturbations.
History is sometimes necessary: In compartmental models with unequal loss rates, the total amount alone cannot close the first-order dynamics; the system requires knowledge of the internal contrast or the history of the total to predict future states.
Joint information is non-additive: Independent sources can provide zero information individually but full information jointly, necessitating linked observations for certain targets.

Significance and Claims
The paper explicitly states its modest scope. It does not claim to establish a new universal law of biological organization or a new reduction algorithm. Its significance lies in providing a critical framework for evaluating descriptions:

It forces the researcher to specify the target response, the domain of inputs, and the tolerance for error before asserting that a description is adequate.
It demonstrates that "response preservation" is a conditional property, not an absolute one. A description adequate for one response (e.g., stationary outflow) may be inadequate for another (e.g., transient retention).
It highlights that empirical adequacy, novelty of application, and predictive or clinical benefit remain to be tested; the calculations in the paper establish model consequences, not empirical facts about living systems.
The paper serves as a conceptual tool for precision medicine and biological organization research (specifically within the author's COSMOS agenda) by clarifying when and why simplified descriptions fail, thereby guiding the selection of discriminating comparisons and measurements.

The author concludes that a description earns its role by resolving a specific question with supported relations and inspectable omissions, rather than by claiming universal validity. The work emphasizes that compatibility between descriptions at different scales is a requirement on the relation between models, not a proven property of nature, and that causal claims require interventions that address competing effects, not just parameter variations within a single model.

More like this