A New Formula for Evolution: How a 60-Year-Old Problem in Genetics Was Solved Using Physics

When a new mutation appears in a population of bacteria, yeast, or cancer cells, its fate is not written in stone. Even if it’s beneficial, it might vanish by chance. Or it might sweep through the population, becoming universal. For decades, evolutionary biologists have relied on a single, elegant formula — derived by Motoo Kimura in 1962 — to predict the odds of this outcome in simple cases. But in the real world, especially among asexual organisms like bacteria, multiple mutations arise at once, competing and interfering with one another. In these complex scenarios, Kimura’s formula breaks down. And until now, no one had found an exact solution.

That changes today. In a breakthrough paper, Emil A. Yuzbashyan, Alexander Feigel, and Alexandre V. Morozov have cracked the problem using an unexpected tool: fluctuation theorems from non-equilibrium statistical physics. These are exact mathematical identities that describe how systems far from equilibrium — like evolving populations — behave over time. By applying this framework, the researchers not only re-derived Kimura’s classic result but also solved the long-standing puzzle of multi-allele fixation probabilities. Their work delivers the first exact, recursive formula for predicting how any new mutation will fare when competing against dozens of others — a capability that could transform how we model evolution in pathogens, tumors, and microbial communities.

The implications are profound. For the first time, scientists can compute the true odds of a drug-resistant mutation taking over a bacterial population without resorting to approximations or simulations. This isn’t just theoretical: it means more accurate forecasts of antibiotic resistance, better models of cancer progression, and deeper insight into how evolution navigates complex genetic landscapes.

The Science

At the heart of evolutionary genetics lies a deceptively simple question: if a single individual in a population of size $N$ acquires a new mutation, what is the probability that this mutation will eventually spread to everyone — that is, become fixed? In 1962, Motoo Kimura answered this for the case of two alleles (e.g., wild-type vs. mutant) using diffusion theory, a mathematical framework that treats allele frequencies as continuous variables evolving under selection, mutation, and genetic drift. His solution, now textbook material, is:

$$\frac{1-e^{-2Nsp}}{1-e^{-2Ns}}$$

where $s$ is the selection coefficient (a measure of fitness advantage), $p$ is the initial frequency of the mutant, and $N$ is the population size. This formula works beautifully — but only when there are just two types in the population.

The real world is messier. In asexual populations — where genes aren’t reshuffled by sex — multiple beneficial and deleterious mutations arise independently and compete. This phenomenon, known as clonal interference, means that a new mutation doesn’t just compete against the wild type; it faces a shifting mosaic of other mutants, each with its own fitness effect. Modeling this requires solving a multi-dimensional version of the backward Kolmogorov equation, a partial differential equation that becomes intractable as the number of alleles increases.

Previous attempts to handle this complexity relied on mean-field approximations — essentially averaging the fitness effects of interfering mutations into a single “effective” background. But these approximations ignore the stochastic dynamics of individual lineages and can mislead. Simulations help, but they’re computationally expensive and don’t offer analytical insight.

The new study takes a radically different approach. Instead of attacking the Kolmogorov equation directly, the authors draw on fluctuation theorems — exact symmetries in the path probabilities of non-equilibrium systems. Originally developed in statistical mechanics to describe entropy production in small systems (like molecular motors), fluctuation theorems have recently found applications in biophysics and information theory. The key insight here is that evolutionary trajectories — sequences of allele frequency changes over time — can be treated as stochastic paths in a high-dimensional space. The probability of a forward path versus its time-reversed counterpart obeys a precise mathematical relationship.

Using a path integral formulation of the Fokker-Planck equation (the continuous-time limit of genetic drift), the authors show that fluctuation theorems impose a strict constraint on fixation probabilities in multi-allele systems. This constraint acts like a conservation law: any valid set of fixation probabilities must satisfy it. Guided by this principle, they construct a recursive analytical solution — a formula that computes the fixation probability of any allele by building up from simpler subsystems.

Their method works across weak, intermediate, and strong selection regimes, and they verify it against both stochastic differential equation solvers and discrete Wright-Fisher simulations. The result is a general, exact framework for multi-allele fixation — something population genetics has lacked for over half a century.

What They Found

The most striking result is not a single number, but a pattern of systematic deviation from classical approximations. When the researchers applied their exact formula to a scenario with $M = 5$ to $25$ alleles — representing a population with many competing mutations — they found that the widely used mean-field Kimura approximation can be off by as much as 25%, depending on the selection strength and number of interfering mutations.

Fixation probability vs. background selection strength

Label	Value
M=5	0.18 %
M=10	0.16 %
M=15	0.14 %
M=20	0.13 %
M=25	0.12 %

Data from Figure 3, averaged over 100 realizations. Fixation probability decreases with more competing alleles.

For example, in a population of $N = 100$ individuals with five alleles, where a new beneficial mutation (with selection coefficient $s_1 = 0.02$) competes against four others with randomly sampled beneficial effects, the exact fixation probability is significantly lower than the Kimura prediction. This gap widens as the background selection pressure increases (measured by $S_{\text{max}}$, the upper bound on interfering mutations’ fitness effects).

The error isn’t random — it’s directional. The mean-field approximation tends to overestimate fixation probabilities because it assumes interference is homogeneous. In reality, clonal interference is heterogeneous: some mutant lineages grow faster, dominate early, and suppress others. The exact solution captures this dynamic competition; the approximation does not.

Another key finding emerges when deleterious mutations are present.

Relative error in mean-field approximation

Label	Value
S_max=0.005	0.08 %
S_max=0.010	0.12 %
S_max=0.015	0.18 %
S_max=0.020	0.22 %

Error grows with background selection strength. Based on Figure 3 and Figure S3.

shows that even when the average fitness of the background is neutral or negative, a beneficial mutation can have a higher fixation probability than predicted — because deleterious lineages die out quickly, reducing competition. This effect vanishes in the mean-field model, which treats the background as static.

Relative error in mean-field approximation

Label	Value
S_max=0.005	0.08 %
S_max=0.010	0.12 %
S_max=0.015	0.18 %
S_max=0.020	0.22 %

Error grows with background selection strength. Based on Figure 3 and Figure S3.

Perhaps most surprisingly, the authors show that their recursive formula remains accurate even when selection coefficients are relatively large (up to $s \approx 0.025$), a regime where perturbative methods typically fail. Beyond that, numerical convergence becomes unreliable — a limitation noted in the paper — but within this range, the solution is robust.

Why This Changes Things

For six decades, Kimura’s formula has been the gold standard in evolutionary genetics. It’s taught in every graduate course, used in hundreds of models, and embedded in software for inferring selection from genomic data. Yet it rests on a simplification: two alleles, no interference.

The new work shatters that limitation. It provides the first exact analytical handle on clonal interference — the very process that governs adaptation in asexual populations. This is not a marginal improvement; it’s a foundational upgrade.

Consider antibiotic resistance. When a bacterial population is exposed to a drug, multiple resistance mutations may arise. Which one fixes — if any — determines whether the infection persists. Current models often assume these mutations compete independently or use simulations to estimate outcomes. Now, for the first time, we can compute the exact probability that a specific resistance allele will dominate, given the full spectrum of competing variants.

The same applies to cancer. Tumors are asexual populations evolving under selection. A new driver mutation doesn’t just compete against normal cells; it faces other mutant clones, some of which may be more aggressive. Predicting which clone will expand — and thus which therapy might fail — requires modeling multi-allele dynamics. The new formula offers a path toward more accurate evolutionary forecasting in oncology.

Beyond applications, this work redefines what’s possible in theoretical biology. It demonstrates that tools from statistical physics — developed for entirely different domains — can solve deep problems in evolution. Fluctuation theorems, once seen as abstract curiosities, are now practical instruments for understanding life’s randomness.

And it challenges a quiet assumption in evolutionary modeling: that approximations are “good enough.” The paper shows they’re not. In

, the relative error between the exact solution and Kimura’s approximation exceeds 20% in many regimes — a difference that could change clinical or ecological predictions.

What’s Next

The immediate impact will be in refining models of asexual evolution. The authors’ recursive formula can replace both simulations and approximate methods in studies of adaptation speed, mutation accumulation, and evolutionary predictability. Because it’s analytical, it allows for sensitivity analysis, parameter inference, and integration into larger frameworks — something simulations cannot easily offer.

But challenges remain. The current solution assumes symmetric mutation rates and constant selection coefficients. Real populations often violate these assumptions: mutation rates vary across sites, selection fluctuates with environment, and fitness effects can be frequency-dependent. Extending the framework to these cases is the next frontier.

Another open question is computational scalability. While the recursion is exact, it grows in complexity with the number of alleles. For genomes with thousands of possible mutations, a full multi-allele treatment may still be infeasible. Dimensionality reduction techniques or hybrid approaches — combining exact solutions for key loci with approximations elsewhere — may be needed.

Perhaps most exciting is the potential for experimental validation. With barcoded microbial lineages and deep sequencing, researchers can now track the fate of dozens of competing clones in real time. The new formula generates precise, testable predictions for fixation probabilities in such experiments — a rare moment when theory outpaces data, daring empiricists to catch up.

In the end, this paper is a reminder that some of science’s hardest problems yield not to brute force, but to cross-pollination. By borrowing a tool from physics — a fluctuation theorem — the authors have given biology a new lens on evolution. We can now see not just the winner, but the full landscape of competition. And in doing so, they’ve brought us closer to predicting, and perhaps even steering, the course of evolution itself.