Best Books to Learn Systems Biology, in Order
Systems biology is the claim that a cell's behaviour lives in the wiring rather than in the parts, and that the wiring can be modelled. That makes it an unusual subject to read into, because it demands two literacies at once — molecular biology and applied mathematics — and most readers arrive with one. This path opens with the argument for the whole enterprise, then spends a stage repairing whichever half you are missing, then works through the standard courses, the modelling toolkit, and finally the network and constraint-based approaches. Be honest with yourself about the prerequisites: from the third stage onward you need differential equations and comfort with linear algebra, and the last stage assumes both plus some probability.
The argument for treating a cell as a system
IntermediateUnderstand why a complete parts list does not explain a living thing, and why causation in biology runs downward as well as upward — the conceptual claim every model in this path is built on.
▸ Study plan for this stage
Pace: Four to six weeks, and this is the only stage on the path with no mathematics in it. Noble's The Music of Life is 176 pages, short and polemical, and reads in two or three evenings at 30-40 pages a sitting; it is the case against genetic reductionism by the physiologist who built the first computati
- Downward causation, and Noble's principle of biological relativity — no privileged level
- Why a complete parts list does not determine behaviour
- The distinction between a gene as a causal agent and a gene as a template used by a system
- Order-of-magnitude estimation as a biological skill
- Characteristic numbers: protein copy numbers, ribosome and polymerase rates, diffusion times across a cell
- Timescale separation, and why knowing which processes are fast decides what a model can ignore
- The difference between a cartoon of a pathway and a model of it
- What does Noble mean by biological relativity, and what would count as evidence against it?
- Why does the heart cell model Noble built serve as his central example rather than a molecular one?
- Roughly how many proteins are in a bacterial cell, and how does that number constrain what a stochastic model must handle?
- How long does a small molecule take to diffuse across a bacterium versus a eukaryotic cell, and what does that difference imply about compartmentalisation?
- Which biological processes are fast enough to be treated as instantaneous relative to gene expression, and how did you decide?
- Work twenty of Milo and Phillips's estimation questions with pen and paper before reading their answers, and record how often you were within a factor of ten
- Estimate, from Milo and Phillips's numbers alone, how long it takes a bacterium to double its proteome, and check the answer against the observed doubling time
- Take Noble's heart cell example and list which levels of organisation his argument requires, then write down what a purely bottom-up account would have to supply instead
- Pick a signalling pathway you know as a cartoon, and annotate every arrow with a rate or a concentration from Milo and Phillips — note how many arrows you cannot annotate at all
- Write half a page on what Noble is and is not claiming, in a form you would be willing to defend against a molecular biologist
Next up: Noble has given you the argument and Milo and Phillips the numbers; the next stage supplies whichever of the two literacies — molecular biology or dynamical systems — you arrived without.
The conceptual opener, by the physiologist who built the first computational model of a heart cell in 1960. Noble's case against genetic reductionism is short, polemical and readable without any mathematics, and it gives you the reason the rest of this path exists.
Noble's fuller and more careful statement of the same argument a decade later, with his principle of biological relativity — that there is no privileged level of causation. Read it second if the first book convinced you and you want the version with the objections answered.
Milo and Phillips's collection of the quantities a modeller needs and biology courses never give you: how many proteins in a cell, how fast a ribosome runs, how long a signal takes to cross a cytoplasm. Read it here because order-of-magnitude intuition is what turns a diagram into a model.
Repairing the half you are missing
IntermediateArrive at the courses with both literacies: enough molecular cell biology to know what is being modelled, and enough dynamical systems to know what a model can do.
▸ Study plan for this stage
Pace: Three to eight months for 2,939 pages, and how long depends entirely on which half you are missing — this stage is deliberately asymmetric and nobody works all three books. Alberts and colleagues' Molecular Biology of the Cell is 1,463 pages and must not be read cover to cover: if you come from phys
- Transcriptional regulation: promoters, transcription factors, operators and cooperative binding
- Signal transduction cascades, and covalent modification as a fast switch
- The cell cycle as a controlled sequence of irreversible transitions
- Fixed points, linear stability analysis and the eigenvalues that classify them
- Phase portraits and nullclines as the working tool for a two-variable system
- Bifurcations — saddle-node, transcritical, pitchfork, Hopf — and what each does to a biological switch or oscillator
- Limit cycles and relaxation oscillations
- Statistical mechanics of binding, and the physical basis of a Hill function
- What makes a transcription factor's binding cooperative, and what does cooperativity do to the shape of the input-output curve?
- Given a two-variable system, how do you find its nullclines and read the qualitative dynamics off them without solving anything?
- What distinguishes a saddle-node bifurcation from a pitchfork, and which one produces hysteresis?
- What has to be true of a two-dimensional system for a limit cycle to exist, and what theorem tells you?
- Where does the Hill function come from physically, and what does the Hill coefficient correspond to in the binding model?
- Which chapters of Alberts does a modeller actually need, and what makes the rest reference material?
- Work Strogatz's exercises on one-dimensional bifurcations until you can sketch the bifurcation diagram of an unfamiliar system without computing anything
- Draw the phase portrait of a two-gene mutual repression system by hand, find the nullclines, and identify the parameter range in which it is bistable
- Derive a Hill function from an equilibrium binding model following Phillips, and confirm the Hill coefficient matches the number of cooperative sites
- Take one signalling pathway from Alberts and write down the smallest set of ordinary differential equations that could represent it, stating every assumption you made to get there
- Reproduce one of Phillips's estimates for a cellular process and compare it with the corresponding number in Milo and Phillips from the previous stage
Next up: You now have both literacies; the next stage is the standard course that assumes them and puts them together on real regulatory circuits.
The reference for the biology, and the book to work the relevant chapters of if you come from physics, maths or engineering — gene regulation, signalling, the cell cycle. Do not read it cover to cover. Our record is an early edition of a text now many editions on, so buy current.
The bridge in the other direction: physics applied to cells, with estimation and simple models throughout, written for people who want the numbers to mean something. The best single preparation for Alon. The record is the first edition; a second exists.
The mathematics prerequisite, and famously the most enjoyable book on differential equations there is: fixed points, stability, bifurcations, oscillators. Every switch, oscillator and bistable circuit in the next stage is one of Strogatz's phase portraits. Buy the second edition rather than our first-edition record.
The standard course
BeginnerAnalyse the recurring circuits of gene regulation — feedforward loops, negative autoregulation, oscillators — and be able to write and solve the equations for a small biological network yourself.
▸ Study plan for this stage
Pace: Five to seven months for 1,221 pages, and this is the core of the path. Alon's An introduction to systems biology is 301 pages and is the most influential book in the field: he builds everything from network motifs, showing that a handful of circuits recur across organisms because they solve recurri
- Network motifs, and the statistical argument that identifies them against a randomised null
- Negative autoregulation and its two effects: response speedup and noise reduction
- The coherent feedforward loop as a sign-sensitive delay, and the incoherent one as a pulse generator
- The repressilator and the requirements for a synthetic oscillator
- Bistability and the toggle switch, and hysteresis as the signature to look for experimentally
- Robustness, and Alon's argument that a design principle is visible in what a circuit is insensitive to
- Quasi-steady-state and nondimensionalisation as the tools that make a model tractable
- Parameter estimation and identifiability — which parameters the data can actually constrain
- Why does negative autoregulation speed up the response time, and what is the tradeoff?
- Explain how a coherent feedforward loop filters a transient input, and derive the delay it introduces
- What has to be true of a two-gene circuit for it to be bistable, and how would you demonstrate hysteresis in the laboratory?
- How is a network motif identified statistically, and what is the null model, and what does the choice of null model change?
- When is the quasi-steady-state approximation valid, and what goes wrong when it is applied outside that regime?
- Which parameters of a small model are typically unidentifiable from time-course data alone, and what experiment would fix that?
- Simulate the coherent and incoherent feedforward loops from Alon and reproduce his sign-sensitive delay and pulse responses
- Nondimensionalise one of Alon's models following Ingalls's procedure and count how many independent parameters actually remain
- Build a toggle switch model, find its bistable parameter region by continuation, and produce the hysteresis curve
- Take Voit's account of how a model is built from data and repeat it on a small published dataset, estimating parameters and reporting which are identifiable
- Do a local sensitivity analysis on the oscillator you built, following Ingalls, and identify which parameter the period depends on most
- Compare Alon's design-principle explanation of one motif with Voit's more empirical treatment of the same circuit and write down what each claims that the other does not
Next up: Alon's models are deterministic and small; the next stage is about choosing the right formalism when they are neither.
The standard course text and the most influential book in the field: Alon builds everything from network motifs, showing that the same handful of circuits recur across organisms because they solve recurring problems. The core of this path. Our record is an early printing; the second edition is substantially revised and is the one now taught.
The gentler and broader alternative to Alon, and the better choice if you want more biology and less design-principle argument: parameter estimation, metabolic and signalling systems, and how a model gets built in practice rather than presented finished.
The most careful treatment of the mathematics itself — nondimensionalisation, quasi-steady-state approximation, sensitivity and bifurcation analysis — worked on biological examples. Read it alongside Alon when a derivation he compresses stops making sense.
The modelling toolkit
BeginnerChoose and implement the right formalism for a given question — deterministic ODE, stochastic simulation, or Bayesian inference from data — and know why molecule numbers decide which one is legitimate.
▸ Study plan for this stage
Pace: Five to seven months for 1,255 pages, and the three books answer three different questions rather than building on each other. Klipp and colleagues' Systems Biology is 504 pages and is the most complete single reference on the methods — kinetic modelling, metabolic control analysis, network analysis
- The chemical master equation, and why it is almost never solvable directly
- The Gillespie stochastic simulation algorithm and its exactness
- Intrinsic versus extrinsic noise, and the experiment that separates them
- When molecule numbers make a deterministic model illegitimate rather than merely approximate
- Metabolic control analysis: flux control coefficients and the summation theorem
- Bayesian inference for kinetic parameters, and likelihood-free methods when the likelihood is intractable
- Model selection and the risk of overfitting a network structure to sparse data
- Robustness, neutral networks and the relationship Wagner argues between tolerance and evolvability
- At what copy number does the deterministic approximation start to mislead, and what quantity determines the threshold?
- How does the Gillespie algorithm sample the next reaction and its time, and in what sense is it exact rather than approximate?
- How would you experimentally distinguish intrinsic from extrinsic noise in gene expression?
- What does a flux control coefficient measure, and what does the summation theorem forbid?
- Why is the likelihood intractable for most stochastic kinetic models, and what does approximate Bayesian computation do about it?
- What is Wagner's argument for why robustness enables rather than prevents evolution, and what evidence does he offer?
- Implement the Gillespie algorithm from Wilkinson for a simple birth-death process and compare the ensemble mean with the deterministic solution as you lower the molecule numbers
- Simulate a genetic toggle switch stochastically and measure the spontaneous switching rate, then compare with the deterministic model, which has none
- Do a metabolic control analysis on a small pathway from Klipp and verify the summation theorem numerically
- Fit a stochastic kinetic model to simulated data using Wilkinson's Bayesian approach, and report posterior uncertainty rather than a point estimate
- Take one of Wagner's robustness arguments and test it on a network you built in the previous stage by perturbing parameters and measuring which outputs survive
- Use Klipp's chapters on standards to export one of your models in a standard exchange format and reload it in a different tool
Next up: Everything so far has been small circuits with known kinetics; the last stage is what to do at genome scale, where you have a network and no parameters at all.
Klipp and colleagues' textbook, and the most complete single reference on the methods: kinetic modelling, metabolic control analysis, network analysis, model fitting and the standards and software the field uses. Encyclopaedic where Alon is argumentative. Note that its subtitled and bare forms are one and the same book.
The necessary correction to the deterministic picture: when a transcription factor is present in tens of copies, averages lie. Wilkinson covers the Gillespie algorithm and Bayesian inference for stochastic kinetic models, with code. Our record is the current third edition.
The evolutionary question the modelling raises and rarely answers: why biological networks tolerate perturbation at all, and how robustness and the capacity to evolve can coexist. Read it as the biological interpretation of everything the models keep showing.
Networks and constraint-based reconstruction
BeginnerWork at genome scale — reconstruct a metabolic network from an annotated genome and analyse it with flux balance analysis — and place biological networks in the wider science of networks.
▸ Study plan for this stage
Pace: Three to four months for 811 pages, and the two books are complementary rather than sequential. Palsson's Systems Biology, Properties of Reconstructed Networks is 336 pages and is the founding text of constraint-based modelling — the route to whole-cell metabolic models when you have no kinetic para
- The stoichiometric matrix, and the null space as the set of feasible steady-state flux distributions
- Flux balance analysis: the linear program, the objective function, and what choosing an objective assumes
- Reconstruction from an annotated genome, and gene-protein-reaction associations
- Gene deletion prediction, essentiality, and where flux balance analysis is known to fail
- Flux variability analysis and the non-uniqueness of the optimal solution
- Degree distributions, scale-free claims, and the ongoing dispute about how well they fit real networks
- Robustness against random failure versus targeted attack, and the hub structure that produces the asymmetry
- Community structure and modularity, and what modularity does and does not imply biologically
- What does the null space of the stoichiometric matrix represent biologically, and why does steady state produce it?
- What does flux balance analysis assume when it maximises growth, and in which organisms and conditions is that assumption known to fail?
- Why is the optimal flux solution usually not unique, and what does flux variability analysis do about it?
- How would you build a gene-protein-reaction association from an annotation, and what happens to the model when the annotation is wrong?
- What is the actual state of the argument about whether biological networks are scale-free, and what does Barabasi say against his critics?
- Which properties of a metabolic network are consequences of biology and which would appear in any large sparse network?
- Load a published genome-scale metabolic model, run flux balance analysis for growth on two carbon sources, and compare the predicted growth rates with published measurements
- Perform an in silico single gene deletion scan on that model and compare predicted essentiality with an experimental essentiality dataset, then examine the disagreements
- Run flux variability analysis on the optimal solution and report how many reactions have genuinely determined fluxes
- Compute the degree distribution of a metabolic network and test the scale-free fit properly rather than by eye, following the cautions in Barabasi
- Remove hubs and then random nodes from the same network and compare how connectivity degrades, reproducing the robustness asymmetry
- Take a module identified by community detection and check whether it corresponds to a known biological pathway or is an artefact of the algorithm
Next up: This is the end of the path: from here the natural continuations are whole-cell modelling, synthetic biology as the engineering application of Alon's design principles, and the single-cell measurement literature that Wilkinson's stochastic models were built to fit.
The founding text of constraint-based modelling, and the route to whole-cell metabolic models when you have no kinetic parameters and never will. Palsson's stoichiometric approach is what industrial metabolic engineering actually runs on. Note our catalogue holds it with a comma in place of the colon.
The general theory of networks — degree distributions, hubs, robustness, community structure — from the person who started the field, with biological networks used throughout as examples. The right closing book, because it shows which properties of a cell's wiring are biological and which are simply what large networks do.
Discussion
Keep reading
Paths that share books, cover the same subject, or open a related topic.
