MATHEMATICS IN BIOLOGY

SPECIAL SECTION

tions provide models of the same general introductory course practical access to con- of computation will provide students with
form. Although the different systems have ceptual tools that are much more sophisti- tools that will serve them well in all of their
important special features (e.g., the conserva- cated than those currently taught in the scientific careers thereafter.
tion laws), surely we would like to commu- standard first-year mathematics courses. Our proposal for an integrated introduc-
nicate the more general idea that dynamics Although real mastery over these ideas will tory education for quantitatively oriented bi-
are described by differential equations and require continuing reinforcement through- ologists really is an experiment in a more
encourage students to discover the applicabil- out the undergraduate curriculum (as is cur- general problem: science education in the
ity of this approach to the dynamics of more rently done for physical science students), a modern world. This is a problem whose so-
complex biological systems through well- unified introduction can empower the stu- lution will require collaborations among
designed laboratory exercises. In a similar dents to explore ideas far beyond what is scientists who now reside in quite different
spirit, statistical physics and kinetic theory currently accessible to them. departments and cultures; enthusiastic as we
provide probabilistic models of the world, but A final and, in the context of biology, are, we also are cognizant of the difficulties
Mendelian genetics is also a probabilistic possibly the most important synergy derives that will no doubt arise. On the other hand,
model and an understanding of probability is from the judicious use of nonstandard exam- the necessary collaborations among the fac-
at the heart of all practical data analysis. ples for basic principles and methods of phys- ulty from several disciplines may well set a
Today, not only can we integrate subjects ics and chemistry. For example, it makes wonderful example for students.
that share common mathematical structures, sense, in modern times, to introduce students To conclude, we believe there is a great
we can also integrate these abstract structures to the idea of molecular motion and thermo- opportunity to construct a unified, mathemat-
with their practical implementation through dynamics in solution rather than focusing ically sophisticated introduction to physics
computation. If the students are taught to only on the world of ideal gases. With afford- and chemistry, which draws on examples
program and to use simple algorithms and if able modern instrumentation, students can from biology wherever possible. Such a
they learn to use high-level languages (e.g., observe and record Brownian motion in a course would provide a coherent introduction
Matlab or Mathematica), they can visualize microscope, for example, and satisfy them- to quantitative thinking about the natural
and verify for themselves the mathematical selves quantitatively how this motion derives world, and invite all students, including biol-
ideas and thereby become comfortable with from invisible molecules bouncing around in ogists of the future, to partake of the grand
those they find less intuitive or more abstract. the solution and even how many such mole- tradition, which flows from Galileos vision.
In statistics, for example, it is possible to cules there must be. This hands-on approach
begin by applying simulation and bootstrap has the advantage that the phenomena (and of
algorithms (e.g., for finding P values). By course the underlying principles) are directly References and Notes
starting in this way, students will more easily and obviously relevant to research in biology. 1. Committee on Undergraduate Biology Education to
Prepare Research Scientists for the 21st Century,
come to appreciate parametric methods and In a similar vein, much of basic combinato- Board on Life Sciences, Division of Earth and Life
closed-form solutions and learn to understand rics, probability theory, and statistics can be Sciences, National Research Council, BIO 2010:
and to use them appropriately. presented in tandem with basic genetics, re- Transforming Undergraduate Education for Future Re-
search Biologists (National Academies Press, Wash-
We believe that integrating mathemat- sulting a substantial saving in overall time ington, DC, 2003).
ics, computation, and the scientific context when compared with separate courses in dif- 2. C. P. Snow, The Two Cultures and the Scientic Rev-
for these ideas will allow students in an ferent departments. Again, the concurrent use olution (Cambridge Univ. Press, New York, 1959).

VIEWPOINT

Uses and Abuses of Mathematics in Biology

Robert M. May

In the physical sciences, mathematical theory and experimental investigation the particulate nature of inheritance were
have always marched together. Mathematics has been less intrusive in the life contemporary with Darwin, and his pub-
sciences, possibly because they have until recently been largely descriptive, lished work accessible to Darwin. Fisher
lacking the invariance principles and fundamental natural constants of physics. and others have suggested that Fleeming
Increasingly in recent decades, however, mathematics has become pervasive in Jenkins fundamental and intractable ob-
biology, taking many different forms: statistics in experimental design; pattern jections to The Origin of Species could
seeking in bioinformatics; models in evolution, ecology, and epidemiology; and have been resolved by Darwin or one of his
much else. I offer an opinionated overview of such usesand abuses. colleagues, if only they had grasped the
mathematical significance of Mendels
Darwin once wrote I have deeply regretted have solved one of Darwins major prob- results (1). But half a century elapsed
that I did not proceed far enough at least to lems. In his day, it was thought that in- before Hardy and Weinberg (H-W) re-
understand something of the great leading heritance blended maternal and paternal solved the difficulties by proving that par-
principles of mathematics; for men thus characteristics. However, as pointed out to ticulate inheritance preserved variation
endowed seem to have an extra sense. Darwin by the engineer Fleeming Jenkin within populations (2).
With the benefit of hindsight, we can see and others, with blending inheritance it is Today, the H-W Law stands as a kind of
how much an extra sense could indeed virtually impossible to preserve the natural Newtons First Law (bodies remain in their
variation within populations that is both state of rest or uniform motion in a straight
Zoology Department, Oxford University, Oxford OX1 observed and essential to his theory of how line, except insofar as acted upon by external
3PS, UK. evolution works. Mendels observations on forces) for evolution: Gene frequencies in a

790 6 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org

 MATHEMATICS IN BIOLOGY

SPECIAL SECTION
population do not alter from generation to explorationand sometimes understanding earlier stages. Various conjectures about under-
generation in the absence of migration, selec- we could not have dreamed of 50 years ago. lying mechanisms can be made explicit in
tion, statistical fluctuation, mutation, etc. Consider the role played by applications of mathematical terms, and the consequences can
Subsequent advances in population genetics, mathematics in sequencing the human and oth- be explored and tested against the observed
led by Fisher, Haldane, and Wright, helped er genomes. This adventure began with the patterns. In this general way, we can, in effect,
make the neo-Darwinian Revolution in the recognition of the doubly helical structure of explore possible worlds. Some hard-nosed ex-
early 20th century. Current work on the one DNA and its implications, an oft-told tale in perimentalists may deride such exploration of
hand provides illuminating metaphors for ex- which classical mathematical physics played a imaginary worlds. And such derision may have
ploring current evolutionary problems, par- central role. Brilliant biochemical advances, al- some justification when the exploration is in
ticularly in molecular evolution, whilst on the lowing the 3 billion-letter-long human se- vaguely verbal terms. The virtue of mathemat-
other hand having important applications in quence to be cut up into manageable fragments, ics in such a context is that it forces clarity and
plant and animal breeding programs (Fig. 1). were a crucial next step. The actual reassem- precision upon the conjecture, thus enabling
Before I embark on chauvinistic elaboration bling of the sequence fragments, to obtain a meaningful comparison between the conse-
of other uses of mathematics in biology today, final human genome sequence, drew on both quences of basic assumptions and the empirical
it is well to reflect on the varied encounters with huge computational power and complex soft- facts. Here mathematics is seen in its quintes-
mathematics that todays nascent biological re- ware, itself involving new mathematics. The sence: no more, but no less, than a way of
searcher is likely to have as undergraduate and sequence information, however, represents only thinking clearly.
earlier. First encounters are usually with A point that arguably deserves more
the simpler aspects of pure mathemat- emphasis than it usually gets is that, in
ics such as numbers, algebra, elemen- such exploration of mathematical mod-
tary trigonometry. Taught well, the em- els, the understanding emerging from
phasis is on clarity of thought, rigor, complex computer-based simulations
even elegance: assume this, prove can often be substantially less complete
that. Taught badly, you get rote learn- than that from the analytic methods of
ing and mechanical tricks. No one taught classical applied mathematics and theo-
well will ever forget the sheer wonder of retical physics. In the World War II
ei 1 (a magical result connecting Manhattan Project, the worlds smartest
two fundamental constants, e and , and theoretical physicists used a mixture of
the ethereal square root of 1). But it can mechanical calculators (the computers
all seem rather abstract, or even off- of their day) and analytic approximation
putting (particularly if taught badly), and techniques to go from basic equations to
for some this will color attitudes to the design of the atom bomb. At every
mathematics for the rest of their lives. stage in this process, there was preserved
In the typical curriculum, applica- an intuitive understanding of the relation
tions come later, often in physics where between the underlying physical as-
the fundamentals are cast in mathemati- sumptions and the results. Today, mete-
cal terms (whence Einsteins reflection orological predictions are based on sat-
on the unreasonable effectiveness of Fig. 1. Difference between the outcomes from blending and ellite data and the Navier-Stokes (N-S)
mathematics in describing the natural from particulate inheritance. In post-Mendelian terms, we hydrological equations, using computers
world). Others will encounter applied assume a single diallelic locus, and hence three diploid geno- whose power is beyond the wildest
mathematics in the form of probability types (AA, blue; Aa, green; aa, yellow). Under particulate imaginings of an earlier generation.
theory and statisticssome through inheritance, the populations variability is preserved from gen- Amongst other things, we now believe
misspent youth playing bridge or poker, eration to generation. In contrast, the conventional wisdom of that the nonlinear N-S equations can
Darwins day saw offspring inherit a blend of parents charac-
others more formally in the design of teristics, here represented as the average of the two parental give so-called chaotic dynamics, where
experiments to separate significant re- shadings. The result is that the variability diminishes in suc- the sensitivity to initial conditions (the
sults from statistical fluctuations and cessive generations (the variance is halved each generation if satellite data) is such that prediction be-
other noise. Among the younger gener- mating is at random). yond 10 to 20 days can be effectively
ation, many will have first met applied impossible, no matter how powerful the
mathematics often without realizing itin the Tycho Brahe stage. Current work on various computer. In many ways, this has greatly in-
computer games. genomes uses pattern-seeking programs to sort creased our understanding of local weather pre-
A paradigmatic account of the uses of math- out coding sequences corresponding to individ- diction. Butan important buta rigorous
ematics in the natural sciences comes, in delib- ual genes from among the background that is mathematical proof that the N-S equations have
erately oversimplified fashion, from the classic thought to be noncoding. Again, elegant and chaotic solutions, as distinct from computer
sequence of Brahe, Kepler, Newton: observed sometimes novel mathematics is involved in simulations that look chaotic, is in effect one of
facts, patterns that give coherence to the obser- this Keplerian stage of the work in progress. the century-old unsolved Hilbert problems (and
vations, fundamental laws that explain the pat- We are only just beginning, if that, the Newton- the Clay Foundation in Boston will give you 1
terns. These days, mathematics enters at every ian stage of addressing the deeper evolutionary million dollars if you solve it).
stage: in designing the experiment, in seeking questions posed by these patterns (not least, the More generally, the increasing speed and
the patterns, in reaching to understand underly- surprising finding of how many genes we share sophistication and ease of use of computers
ing mechanisms. In biology, of course, every with other species, and how numbers of genes enables an increasingly large number of life
stage in this caricature is usually vastly more appear to be uncorrelated with what we regard as scientists who have no substantial background
complex than in the early days of physics. But complexity of the organism; rice, for example, in mathematics to explore mathematical mod-
the advent of computers, and the extraordinary appears to have more genes than we do). els and draw conclusions about them. Such
doubling of their capability roughly every 18 In this Newtonian quest, mathematical mod- activity usually consists of representing sensible
months for the past several decades, permits els will help in a different way than in the and evidence-based assumptions as the starting

www.sciencemag.org SCIENCE VOL 303 6 FEBRUARY 2004 791

 MATHEMATICS IN BIOLOGY
SPECIAL SECTION

point for a complicated and usually nonlinear the human food chain, after the precautionary would conclude that for every case of vCJD
dynamical system, assigning particular param- measures taken earlier; the amount of infected prevented by banning BOTB, you were looking
eters (often in an arbitrary way), and then let- meat additionally removed by banning BOTB; at 100,000 (give or take a factor 10!) cases of
ting this complicated system rip. This repre- the probability, P, of contracting vCJD after vCJD yet to come from the earlier epidemic.
sents a revolutionary change in such theoretical eating, say, 1 g of infected meat. Of these three, Put another way, given the best guess of a few
studies. Until only a decade or two ago, anyone the first can be estimated fairly reliably (cer- hundred cases of vCJD from the earlier epidem-
pursuing this kind of activity had to have a solid tainly to within a factor 10). The second can be ic, the ban prevented 103 of a case. The public,
grounding in mathematics. And that meant that estimated to stupefying precision, which indeed with admirable common sense, saw the ban as
such studies were done by people who had forms the bulk of the report. The infection silly, and they were right. Sadly, examples of the
some idea, at an intuitive level, of how the probability, P, is essentially unknown. So the application of statistical confidence intervals to
original assumptions related to the emerging study ran computer simulations, with an arbi- distributions resulting from making arbitrary as-
graphical display or other conclusions on their trary distribution of P values ranging over six sumptions about essentially unknown parameters,
computer. Removing this link means that we orders of magnitude. For any one assumption and then endowing this with reality by passage
arguably are seeing an increasingly large body about the P value, you could roughly assess the through a computer, continue to proliferate.
of work in which sweeping conclusions number of vCJD cases saved by banning BOTB The history of useful mathematical models,
emergent phenomenaare drawn from the by a calculation on the back of an envelope. In and of mathematics more generally, varies
alleged working of a mathematical model, without the published study, however, these arbitrary among different areas in the life sciences. I
clear understanding of what is actually going referred earlier to population genetics;
on (Fig. 2). I think this can be worrying. ecology and immunology provide two fur-
This cautionary note can be fleshed out ther, and interestingly different, examples.
with a couple of examples. The first Ecology is a relatively young subject,
comes from HIV/AIDS in the mid-1980s, and much early work was largely descrip-
when Roy Anderson and I published the tive. Seminal studies by Lotka and Volt-
first, rough estimate of its likely demo- erra explored mathematical metaphors for
graphic impact in some central African competition and other interactions among
countries. The main unknown at that time species, but things did not really take off
was the probability, , that an infected until the 1960s, when Hutchinson and
individual would infect a susceptible part- MacArthur began to ask focused and test-
ner. Available data suggested that de- able questions in the idiom of theoretical
pended relatively little on the number of physics: How similar can species be yet
sexual acts within a partnership. On this persist together. How do the patterns of
basis, we used a relatively simple model species interactions within a food web
to suggest that the future demographic affect its ability to withstand disturbance.
impact of HIV/AIDS could be severe in Why are some natural populations rela-
some such countries. In contrast, the Fig. 2. The predictions for the numbers of animals infected tively steady from year to year, others
with foot-and-mouth disease, under a controlled program
World Health Organization and the Pop- that combines culling with reactive vaccination. The red cyclic, and others fluctuate widely (5)? At
ulation Council in New York produced curve shows the mean, and the shaded region the range, for first, some ecological empiricists resented
models that were much more complex, simulations of a detailed stochastic model in which spread arrivistes, who had paid no dues of years
including very detailed demographic data, of infection is modeled using the actual spatial distribution of toil in the field, presuming to math-
but where HIV transmission probability of farms in England and Wales. The blue curve is from a ematicize their problems (sometimes
was treated as if for measles, compound- highly simplied toy model, using gross averages of sweeping aside arguably irrelevant, but
relevant parameters. The comparison between the simple
ing independently randomly for each in- model, in which the dynamics can be clearly understood, certainly much loved, details in the pro-
dividual sex act. Thus, in effect, their and the complex computer simulations is illuminating both cess). Others welcomed the newcomers
models assumed that, knowing nothing of for the similarities and the differences (the latter helping to too uncritically. Look at the ecology texts
the infective status of individuals, 1 sex explain why, in reality, local outbreaks continued to erupt of 50 years ago, and you will find very
act with each of 10 different sex workers many months after the endpoint predicted by deterministic few equations. Todays, by contrast, con-
was effectively equivalent to 10 acts with models). Reprinted with permission (10). tain a blend of observation, field and lab-
1; our data-governed, but otherwise much oratory experiment, and theory expressed
simpler, model saw the former as roughly 10 assumptions were magically transmogrified in mathematical terms. I think this reflects the
times more risky. So it was not surprising that into a probability distribution by passage maturing state of this vital subject, although it
the later models, apparently more realistic by through a big computer. The resulting 95% still has more questions than answers. The
virtue of their computational complexity, sug- confidence interval included the possibility of mathematical traffic, moreover, has not been all
gested a less gloomy view than ours. Sadly, but saving one life. This, along with other consid- one-way: Some of the seminal developments in
understandably, our predictions have proved erations, led to the ban. Of course, an alterna- chaos theory were prompted by ecological
more reliable (3, 4). tive assessment could have asked what is the problems (6, 7).
An interestingly different but rather com- ratio of the cases of vCJD saved by banning Immunology offers a somewhat different
mon sin is exemplified in a risk-assessment BOTB to the number of casesat that time picture. Here there are truly remarkable advanc-
study commissioned by the U.K. Government a hard to estimate coming down the pipeline es in describing and understanding, at the mo-
few years ago to help determine whether to ban from cattle that entered the human food chain at lecular level, how individual viruses and other
beef-on-the-bone (T-bone steak; hereafter the height of the epidemic, before effective infectious agents interact with individual im-
BOTB) in the aftermath of the BSE or mad measures were taken. This ratio depends mainly mune system cells. And on the basis of such
cow epidemic. Three factors are involved in on the number of cattle still entering the human knowledge, so brilliantly detailed on the molec-
any assessment of the likely number of cases of food chain to the number entering at the epi- ular scale as almost to defy intuitive compre-
vCJD in humans prevented by such a ban: the demics height, with the great uncertainty in P hension, we can, for example, design drugs that
number of infected animals still liable to enter canceling out of the ratio. On this basis, you suppress viral replication. Chemotherapy

792 6 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org

 MATHEMATICS IN BIOLOGY

SPECIAL SECTION
against HIV is one notable example. At the edge the usefulness of statistical recipes to help the problem are misty or a vital parameter is
same time, however, there is as yet no agreed design and analyze experiments. More familiar uncertain to within, at best, an order of magni-
explanation for why there is so long, and so in some areas than others are the benefits of tude. It makes no sense to convey a beguiling
variable, an interval between infection with mathematical studies that underpin pattern- sense of reality with irrelevant detail, when
HIV and onset of AIDS. Indeed, I guess that seeking and other software that is indispensable other equally important factors can only be
many researchers in this field do not even think in elucidating genomes, and ultimately in un- guessed at. Above all, remember Einsteins dic-
about this question. But I suspect the answer derstanding how living things assemble them- tum: models should be as simple as possible,
may necessarily involve understanding how selves. Very generally useful are still-unfolding but not more so.
whole populations of different strains of HIV advances that illuminate the frequently counter-
interact with whole populations of different intuitive behavior of nonlinear dynamical sys-
References and Notes
kinds of immune system cells, within infected tems of many kinds. 1. R. A. Fisher, The Genetical Theory of Natural Selection
individuals. And understanding the nonlinear Mathematics, however, does not have the (Dover [reprint], New York, 1958).
dynamics of such a system will require mathe- long-standing relation to the life sciences that it 2. G. H. Hardy, Science 28, 49 (1908).
3. For a more full discussion see R. M. Anderson and
matical models with similarities to and differ- does to the physical sciences and engineering. It R. M. May [Infectious Diseases of Humans (Oxford
ences from those that have helped us under- is therefore not surprising to find occasional Univ. Press, 1991), chap. 11].
stand population-level problems in ecology and abuses. Some have been sketched above. Par- 4. J. Bongaarts, Stat. Med. 8, 103 (1989).
5. R. MacArthur, Geographical Ecology (Harper & Row,
infectious diseases (8). It may even be that the ticularly tricky are instances in which con- New York, 1972).
design of effective vaccines against protean ventional statistical packages (often based on 6. For further discussion, see G. Farmelo (Ed.) [It Must
agents like HIV or malaria will require such assumptions of an underlying Gaussian distri- be Beautiful: Great Equations of Modern Science
population-level understanding. As yet, this butionthe central limit theorem) are applied (Granta, London, 2002), pp. 212229] and (7).
7. J. Gleick, Chaos: Making a New Science (Viking, New
mathematically theoretical aspect of immunol- to situations involving highly nonlinear dynam- York, 1987).
ogy is even less to be found in textbooks than ical processes (which can often lead to situa- 8. M. A. Nowak, R. M. May, Virus Dynamics: Mathemat-
were mathematical models in ecology texts a tions in which rare events are significantly ical Principles of Immunology and Virology (Oxford
Univ. Press, Oxford, 2000).
generation ago. I venture to predict that the more common than Gaussian distributions sug- 9. N. Goldeneld, L. P. Kadanoff, Science 284, 87
corresponding immunology texts will indeed gest) (9). Perhaps most common among abuses, (1999).
look different in 20, or even 10, years time. and not always easy to recognize, are situations 10. M. J. Keeling et al., Nature 421, 136 (2003).
In short, mathematical models have proved where mathematical models are constructed 11. I thank many colleagues, particularly S. Levin, A.
Grafen, D. Rubenstein, and M. Stumpf for helpful
to have many uses and to take many forms in with an excruciating abundance of detail in conversations, and the Leverhulme Trust and the
the life sciences. We all, by this time, acknowl- some aspects, whilst other important facets of Nufeld Foundation for their support.

REVIEW

Evolutionary Dynamics of Biological Games

Martin A. Nowak1* and Karl Sigmund2,3

Darwinian dynamics based on mutation and selection form the core of mathe- can be part of the fitness landscape of the
matical models for adaptation and coevolution of biological populations. The other. A hosts successful immune response
evolutionary outcome is often not a tness-maximizing equilibrium but can to a pathogen, for instance, will exert se-
include oscillations and chaos. For studying frequency-dependent selection, lection pressure leading to adapted strains
game-theoretic arguments are more appropriate than optimization algorithms. of pathogens, and vice versa (35). But
Replicator and adaptive dynamics describe short- and long-term evolution in even within a single population, the fitness
phenotype space and have found applications ranging from animal behavior and of a trait often depends on the prevalence of
ecology to speciation, macroevolution, and human language. Evolutionary game that trait: The selective advantage of a giv-
theory is an essential component of a mathematical and computational approach en tree height, for example, depends on the
to biology. heights of neighboring trees. Similarly, the
success of a given sex ratio depends on the
Evolution through natural selection is often population would therefore be expected to overall sex ratio in the population.
understood to imply improvement and increase over time. This is often pictured as Therefore, the fitness landscape is
progress. A heritable trait that confers to its a steady ascent on a so-called fitness land- shaped by the phenotypic distributions of
bearer a higher fitness will spread within scape. The landscape metaphor suggests the involved populations. As the population
the population. The average fitness of the some solid ground over which the popula- moves through the fitness landscape, new
tion moves. This paradigm (1), which is peaks and valleys form, channeling its fur-
1
Program for Evolutionary Dynamics, Department also widespread in the theory of genetic ther motion. This viewpoint affects not
of Mathematics, Department of Organismic and algorithms (2), neglects one-half of the only the intuition of evolutionary biologists
Evolutionary Biology, Harvard University, 1 Brattle evolutionary mechanism: Although the en- but also their theoretical tools. The proper
Square, Cambridge, MA 02138, USA. 2Faculty for vironment selects the adaptations, these ad- technique for describing uphill motion on
Mathematics, University of Vienna, Nordberggasse
15, A-1090 Vienna, Austria. 3International Institute
aptations can shape the environment. By solid ground is optimization theory, a set of
for Applied Systems Analysis (IIASA), A-2361 Lax- moving across a fitness landscape, popula- mathematical techniques developed in the
enburg, Austria. tions change that landscape (Fig. 1). past 300 years, mostly to solve physical or
*To whom correspondence should be addressed. This is particularly clear if several pop- technical problems. If the adaptive steps,
E-mail: [email protected] ulations interact, because each population however, imply changes in the environ-

www.sciencemag.org SCIENCE VOL 303 6 FEBRUARY 2004 793