How Science Works | Mathematical Biology — Equations, Networks, Populations, Dynamics and Models of Living Systems

Safety and scope: examples are educational mathematical models. They do not provide harmful biological optimisation, clinical recommendations or operational pathogen-engineering procedures.

Reading route: build the model → follow dynamics → model populations and ecology → model cells and molecules → add space and stochasticity → connect model to evidence.

1. The scientific job begins by deciding what to leave out

Every biological system contains more detail than any tractable model can include.

Mathematical biology begins by choosing state variables, interactions and timescales that preserve the mechanism relevant to the question while deliberately ignoring detail that should not affect the answer.

2. A variable is a biological claim in compressed form

Population size, protein concentration, infection prevalence, membrane voltage or tumour volume can each be represented as variables.

Choosing a variable asserts that this quantity is measurable or meaningful enough to represent part of the system state.

3. Parameters describe how the model behaves

Growth rates, diffusion coefficients, carrying capacities, reaction constants and interaction strengths are parameters.

Some are directly measurable, some estimated from data and some effective summaries of many hidden processes.

4. Units expose impossible equations

Terms added together must have compatible dimensions.

Dimensional checking catches many modelling mistakes before any numerical solver is run.

5. Worked example: units can diagnose a missing timescale

Original example. If population N is measured in individuals and dN/dt in individuals per day, a term rN must also have individuals per day.

Therefore r must have units of 1/day. A parameter written as “0.2 individuals” cannot serve as the growth-rate constant in that equation.

6. Mechanistic and phenomenological models answer different questions

A mechanistic model encodes proposed causal processes. A phenomenological model summarises observed patterns without necessarily representing underlying mechanism.

Both can be useful if their claims are kept distinct.

7. Continuous models treat change as smoothly varying

Differential equations are natural when large populations or concentrations change continuously enough that individual events can be averaged.

They trade microscopic detail for analytic structure and computational efficiency.

8. Discrete models preserve event steps

Difference equations, branching processes, cellular automata and individual-based models update systems in distinct events or time steps.

They can represent births, contacts, cell divisions or movement more explicitly than a continuous average.

9. Deterministic models give the same trajectory from the same initial state

If initial conditions and parameters are fixed, deterministic equations return one trajectory.

This is useful for mean behaviour but can hide biological variability and chance events.

10. Stochastic models make chance part of the mechanism

Births, deaths, mutations, gene-expression events and encounters occur discretely and with randomness.

Stochastic models represent distributions of possible outcomes rather than one path.

11. Initial conditions can determine which future is reached

Nonlinear biological systems can contain multiple stable states.

Two systems with the same parameters can therefore evolve differently because they begin on different sides of a threshold.

12. Nondimensionalisation reveals the ratios that actually matter

Variables and time can be rescaled so equations depend on fewer dimensionless parameter groups.

This can expose that two apparently different biological systems obey the same mathematical regime because key ratios are similar.

13. Exponential growth is a local model, not an eternal law

The equation dN/dt = rN says each individual contributes proportionally to future growth while resources and conditions remain effectively unconstrained.

It is useful over limited ranges and fails when density-dependent effects matter.

14. Worked example: doubling time comes from the growth rate

Original calculation. For exponential growth N(t)=N₀e^{rt}, doubling requires e^{rt}=2.

So t = ln2/r. If r = 0.1 per day, doubling time is about 6.93 days.

15. Logistic growth adds density dependence

The logistic model multiplies growth by a factor such as 1 − N/K, reducing net growth as population N approaches carrying capacity K.

K is not a mystical fixed number; it summarises environmental constraints under the conditions represented.

16. Equilibria are states where modelled change stops

An equilibrium satisfies the equations with zero rate of change.

The system can still contain active microscopic processes; equilibrium only means the chosen state variables remain constant in the model.

17. Stability asks what happens after a small disturbance

A stable equilibrium pulls nearby trajectories back. An unstable equilibrium pushes them away.

This distinguishes merely possible states from states the biological system can maintain.

18. Linearisation turns local nonlinear behaviour into a tractable approximation

Near an equilibrium, nonlinear equations can often be approximated by their Jacobian matrix.

Eigenvalues then reveal whether small perturbations decay, grow or oscillate locally.

19. Oscillations can emerge from interaction and delay

Feedback loops, predator–prey interactions and biological clocks can generate repeated cycles without an external periodic driver.

Mathematics separates sustained oscillation from damped transients and noise-driven fluctuations.

20. Bifurcations mark qualitative changes in system behaviour

A gradual parameter change can create or destroy equilibria, trigger oscillations or alter stability.

Biological transitions can therefore be abrupt even when environmental forcing changes smoothly.

21. Worked example: a threshold can turn extinction into persistence

Original conceptual model. Imagine a population whose per-capita growth is negative below a minimum density because individuals rarely find mates.

Above the threshold, growth becomes positive. The same environment can therefore lead to extinction or persistence depending on initial population size.

22. Delay equations remember the past explicitly

Development, maturation and transcription often take finite time.

Delay differential equations let current change depend on earlier states, which can create instability or oscillation absent from an instantaneous model.

23. Difference equations can behave very differently from continuous equations

When reproduction occurs in seasonal generations, discrete updates may be more natural.

Strong nonlinear feedback in discrete maps can produce cycles or chaotic behaviour even with simple rules.

24. Chaos is deterministic unpredictability

Chaotic systems obey deterministic equations yet show sensitive dependence on initial conditions.

Long-term exact prediction becomes impossible even if the model is perfectly known, while statistical properties may remain predictable.

25. Predator–prey models turn interaction into coupled equations

Prey abundance affects predator growth; predators affect prey decline.

Coupling the equations reveals behaviours that cannot be inferred from either population alone.

26. Competition models expose coexistence conditions

Two species can limit one another while also limiting themselves.

Relative strengths of intra- and interspecific competition determine whether one excludes the other or stable coexistence is possible.

27. Mutualism can amplify benefit and instability together

Positive interactions can increase growth of both partners.

Simple mutualism equations can become unrealistically explosive unless saturation or resource limits are included, showing how mathematics exposes missing biology.

28. Age-structured models remember who is in the population

Individuals of different age or stage can have different survival and reproduction.

Matrix population models track movement among classes and identify long-term growth rates and stable stage distributions.

29. Worked example: total population can hide reproductive structure

Original reasoning example. Two populations each contain 1,000 individuals.

If one contains mostly juveniles approaching reproductive age while the other contains mostly old individuals, their short-term growth can differ sharply despite equal totals.

30. Metapopulation models connect local extinction to regional persistence

Habitat patches can lose local populations while dispersal recolonises empty patches.

Regional persistence depends on the network of colonisation, extinction and connectivity rather than one patch alone.

31. Evolutionary game theory models frequency-dependent success

The fitness of a strategy can depend on what others in the population do.

Payoff matrices and replicator equations make this dependence explicit without assuming conscious calculation by organisms.

32. Adaptive dynamics follow traits as ecological conditions change

Mutant invasion success can depend on the resident population’s ecological equilibrium.

This couples evolutionary change to the environment organisms themselves create.

33. Epidemiological compartment models are population models

Susceptible, infectious, recovered or other categories can be represented as compartments connected by transition rates.

The model is useful for population-level dynamics but does not replace individual diagnosis or clinical care.

34. Contact structure changes epidemic behaviour

Homogeneous mixing assumes everyone contacts everyone else proportionally.

Age, household, workplace, geography and social networks break that assumption and can change thresholds and intervention effects.

35. Network models preserve who can interact with whom

Nodes represent individuals or units and edges represent potential interaction.

Degree distributions, clustering and communities can strongly alter spreading, synchronisation and robustness.

36. Percolation models ask when connectivity suddenly spans a system

As links or occupied sites accumulate, a network can cross a threshold where a giant connected component emerges.

The concept appears in ecology, epidemiology and tissue connectivity because global behaviour can emerge abruptly from local links.

37. Enzyme kinetics turn molecular interactions into rate equations

Michaelis–Menten-style models compress binding and catalysis into relationships between substrate concentration and reaction rate under stated assumptions.

The familiar hyperbola is a model consequence, not a universal law for every enzyme system.

38. Worked example: saturation creates diminishing returns

Original example. In v = Vmax S/(Km + S), increasing S from Km to 9Km raises rate from 0.5Vmax to 0.9Vmax.

Nine times more substrate than Km does not produce nine times the half-maximal rate because enzyme capacity saturates.

39. Gene-regulatory models encode feedback and thresholds

Production and degradation terms can be combined with nonlinear regulation functions.

These models can create switching, memory and oscillation from molecular feedback.

40. Boolean networks trade concentration detail for logical structure

Genes or proteins can be represented as on or off with update rules determined by regulators.

The simplification is valuable when network topology is known better than kinetic parameters.

41. Metabolic flux models use conservation constraints

Stoichiometric matrices encode how metabolites are produced and consumed.

Flux-balance methods identify feasible steady-state reaction patterns, while objective functions or extra data choose among many possibilities.

42. Electrophysiological models translate ion currents into voltage dynamics

Membrane voltage changes as ionic currents charge or discharge membrane capacitance.

Conductance-based models show how nonlinear ion-channel feedback can generate action potentials and repetitive firing.

43. Biomechanical models link force to living motion

Differential equations describe limb dynamics, tissue deformation, fluid interaction and muscle–tendon mechanics.

The mathematics complements Biomechanics, which retains physical interpretation and measurement of living movement.

44. Physiological control models formalise homeostasis

Feedback equations can represent temperature regulation, glucose control or cardiovascular responses at an abstract system level.

Clinical application requires validated medical models and qualified interpretation beyond this educational route.

45. Diffusion equations describe spreading down concentration gradients

Fickian diffusion produces partial differential equations in which local concentration changes depend on spatial curvature.

Diffusion sets characteristic timescales that grow with the square of distance.

46. Worked example: diffusion time rises rapidly with distance

Original scaling example. If characteristic diffusion time scales as L²/D, doubling distance L multiplies time by four when D is unchanged.

This is why diffusion is effective across cellular dimensions but inefficient for long-distance transport in large organisms without circulatory systems.

47. Reaction–diffusion systems can create spatial patterns

Local chemical reactions combined with different diffusion rates can destabilise a uniform state and create spots, stripes or waves.

The mathematics shows how pattern can emerge from initially smooth conditions.

48. Advection adds transport by bulk flow

Blood flow, wind, ocean currents and cytoplasmic transport move substances in addition to diffusion.

Advection–diffusion equations compare directed transport with random spreading.

49. Individual-based models preserve heterogeneity

Each simulated organism or cell can carry its own state, location and behavioural rules.

Population patterns then emerge from many local interactions rather than being imposed as averaged equations.

50. Agent-based detail has a cost

More detailed individual rules create more parameters, more computation and more ways to fit data.

A complex simulation is not automatically more explanatory than a simpler model.

51. Branching processes capture early stochastic growth

When populations are small, whether each individual leaves zero, one or several descendants can determine extinction or expansion.

Expected growth may be positive while extinction remains possible by chance.

52. Random walks model movement and search

Successive movement steps with probabilistic directions can approximate animal foraging, molecular motion or cellular migration.

Different step-length and directional rules produce different large-scale dispersal patterns.

53. First-passage problems ask when a stochastic process reaches a target

A molecule finding a binding site, an animal reaching habitat or a population crossing a threshold can all be posed as first-passage questions.

Often the distribution of arrival times matters more than the average alone.

54. Parameter estimation works backward from data

A forward model maps parameters to predictions. Estimation asks which parameter values make predictions consistent with observed data.

Noise and model mismatch mean the answer is usually a distribution or uncertainty region rather than one exact number.

55. Identifiability asks whether the data can distinguish parameters

Two different parameter combinations can sometimes generate indistinguishable outputs.

No optimisation algorithm can recover uniquely what the experiment does not identify.

56. Sensitivity analysis shows which assumptions control predictions

Parameters are varied and changes in outputs are measured.

High sensitivity marks variables that strongly influence prediction; low sensitivity can mark robustness or simply a poorly chosen experimental regime.

57. Uncertainty propagates through nonlinear models

Small uncertainty in inputs can become asymmetric or amplified in outputs when equations are nonlinear.

Prediction intervals should therefore be computed from the model rather than guessed by attaching a symmetric error bar.

58. Model comparison asks which simplification earns its complexity

A larger model can almost always fit old data at least as well.

Held-out prediction, information criteria, perturbation experiments and biological interpretability help determine whether extra mechanisms are justified.

59. Validation should use conditions not consumed during fitting

A model calibrated only on one environment can appear excellent while failing under a new perturbation.

Successful prediction under new conditions is stronger evidence that the encoded mechanism travels.

60. Models can be useful while literally false in detail

A logistic growth equation does not claim every individual senses total population density directly.

It can still capture a useful aggregate constraint if density-dependent effects combine approximately into the chosen form.

61. Common mathematical-biology failure modes

• Equation equals mechanism: forgetting that equations are representations built from assumptions.
• Best fit equals true parameters: ignoring identifiability.
• More detail equals more realism: ignoring overfitting and unmeasured rules.
• Average equals individual: erasing stochasticity and heterogeneity.
• Equilibrium equals inactivity: ignoring balanced internal fluxes.
• Prediction outside the fitted regime is free: ignoring extrapolation risk.

62. How to think like a mathematical biologist

State the biological question first. Choose the smallest set of variables that can express the proposed mechanism. Check units. Identify assumptions and timescales. Analyse equilibria, stability and thresholds before simulating. Ask which parameters can actually be learned from available data. Make predictions that a new experiment could falsify.

63. A staged learning route

First encounter: graphs, rates, exponential growth, probability and simple feedback.

Secondary-to-JC bridge: logistic growth, coupled equations, matrices, diffusion, networks and stochastic processes.

Higher resolution: dynamical systems, bifurcation theory, stochastic differential equations, reaction–diffusion systems, inverse problems, identifiability and multiscale models.

64. Checkpoints with answers

Why is exponential growth not a universal population law? It assumes per-capita growth remains constant and resources or density effects do not yet dominate.

Can two parameter sets fit the same biological data? Yes. Non-identifiability can make different parameter combinations observationally indistinguishable.

Why use a stochastic model? When discrete random events or small populations create variability that an average deterministic trajectory would hide.

What makes a mathematical model scientific? Explicit assumptions, measurable variables, testable predictions and willingness to fail against data.

65. The final skill is knowing which mathematics the biology has earned

A mathematical model should be neither simpler nor more complicated than the biological question can support. The strongest models compress reality enough to reason with while preserving the mechanism needed for prediction—and they make their losses visible.

Sources and connected subjects

Useful foundations include the Society for Mathematical Biology, whose research community spans mathematical and computational biology, together with modern dynamical-systems, ecological, epidemiological and cellular-modelling literature. Worked examples above are original teaching constructions.

Continue to Systems Biology, Ecology, Epidemiology and Statistics.

Return to How Science Works or the How X Works Hub.