📄 evolutionary biology

Emergence of mutants in bacterial populations: a simulation-based approach for parameter inference in extended Luria & Delbru&x0308;ck scenarios

This paper introduces a stochastic simulation framework combined with Approximate Bayesian Computation to accurately infer bacterial mutation rates and fitness effects in complex scenarios where traditional mathematical models fail due to restrictive assumptions like cell death and variable growth rates.

Original authors: Tauzin, A., Frenoy, A.

Published 2026-09-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tauzin, A., Frenoy, A.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

Life is built on a constant, quiet tension between stability and change. Every time a bacterium divides to make a copy of itself, it must replicate its genetic code, a complex instruction manual for building a living cell. This copying process is rarely perfect; occasionally, a letter is misspelled or a word is skipped. These errors, known as mutations, are the raw material of evolution. Most of them are harmless or even harmful, causing the cell to function poorly or die. But sometimes, a mutation provides a new advantage, allowing the organism to survive a harsh environment or resist a poison. To understand how life adapts, scientists must know how often these errors happen. This rate of change is a fundamental number in biology, but measuring it is surprisingly difficult because the events are rare and the outcomes are chaotic.

For decades, scientists have relied on a classic experiment to count these rare errors. The method, known as a fluctuation test, involves growing many separate cultures of bacteria in parallel. Because mutations happen randomly during the growth of each culture, the number of mutant cells in each tube varies wildly. Some tubes might have no mutants at all, while others might have hundreds, depending on when the first error occurred. By counting the mutants in these tubes, researchers can work backward to calculate the mutation rate. However, the mathematical formulas used to solve this puzzle were built on a set of strict assumptions. They assumed that all bacteria grow at the same speed, that no cells die during the experiment, and that every single mutant is found when the culture is plated onto a dish. In the real world, these assumptions often fail. Bacteria die under stress, some mutations make the cells grow slower or faster than their neighbors, and scientists often cannot plate the entire culture, only a fraction of it. When these real-world complications occur, the old formulas break down, leading to incorrect answers about how fast evolution is happening.

To solve this problem, a team of researchers in France has developed a new way to calculate mutation rates that does not rely on rigid formulas. Instead of trying to force complex biological realities into a simple equation, they built a digital simulation of the entire experiment. Imagine a computer program that acts as a virtual laboratory, where millions of virtual bacteria are born, grow, die, and mutate according to specific rules. The researchers programmed this simulator to handle the messy details that the old math ignored: cells dying at different rates, mutations that change how fast a cell grows, and the fact that scientists often only look at a sample of the culture rather than the whole thing. The simulator runs the experiment thousands of times, creating a vast library of possible outcomes for any given set of conditions.

The core of their new method is a process of matching. The researchers take the actual data from a real experiment—the number of mutant bacteria found in a set of test tubes—and compare it against the millions of outcomes generated by their simulator. They adjust the numbers in the simulator, such as the mutation rate or the death rate, and run the simulation again, looking for the set of parameters that produces a pattern of results most similar to the real data. This approach, which uses a statistical technique to find the best fit without needing a traditional formula, allows them to infer the mutation rate even when the conditions are far from perfect. They tested this method against the best existing tools used by scientists today. In simple situations where the old formulas work, their new method performed just as well, proving it was accurate. But when they introduced complications like cell death or mutations that slowed down growth, the old tools failed or gave wildly inaccurate results, while their simulation-based approach continued to find the correct answer.

The researchers also discovered the limits of what can be known. They found that while their method could accurately estimate the mutation rate and the fitness effect of a mutation at the same time, it could not reliably separate the mutation rate from the death rate if both were unknown. The data simply did not contain enough information to tell the difference between a high mutation rate with low death and a low mutation rate with high death. This is a crucial finding, as it tells scientists that to understand these complex scenarios, they must measure cell death in a separate experiment rather than trying to guess it from the mutation data alone. The new method also handles situations where the environment changes during the experiment, such as a culture that starts under stress and then recovers, a scenario that no other tool could currently analyze.

This work represents a shift in how scientists approach complex biological questions. By replacing a single, fragile equation with a flexible, computer-driven model, the researchers have opened the door to studying evolution in conditions that more closely resemble the real world. Their tool allows scientists to ask questions about bacterial survival in the presence of antibiotics, in crowded environments, or under nutritional stress, without having to worry that the mathematical model is forcing the data into a shape it does not fit. The method is not a magic bullet that solves every problem instantly; it requires more computing power and time than the old formulas. However, for the first time, researchers have a reliable way to measure the speed of evolution when the rules of the game are complicated, ensuring that our understanding of how life adapts is based on reality rather than on convenient assumptions.

Technical Summary: Emergence of Mutants in Bacterial Populations: A Simulation-Based Approach for Parameter Inference in Extended Luria & Delbrück Scenarios

Problem Statement
Traditional methods for inferring bacterial mutation rates rely on fluctuation assays analyzed through analytical mathematical models (e.g., the Ma-Sandri-Sarkar maximum likelihood estimator, MSS-MLE). These models depend on restrictive assumptions that are frequently violated in real-world biological settings. Key assumptions include equal growth rates for wild-type and mutant populations, negligible cell death, constant death rates, and complete detection of all mutants (no partial plating). While refined analytical tools (such as rSalvador, Flan, and mlemur) have been developed to address specific deviations like differential fitness or constant mortality, they fail to handle complex scenarios involving simultaneous cell death, unknown fitness effects, and non-constant (e.g., piecewise) death rates. Furthermore, no analytical formula exists for the expected distribution of mutants under these combined complex conditions.

Methodology
The authors propose a simulation-based inference framework to overcome the limitations of analytical models. The approach consists of two main components:

Forward Stochastic Simulator: The authors developed a fast, hybrid simulator for an extended "Birth-Death-Mutation-Selection-Sampling" model.
- Mechanism: It approximates wild-type birth and death using deterministic techniques while simulating mutation events and mutant birth/death stochastically.
- Parameters: The model incorporates mutation rate ( $\mu$ ), relative fitness of the mutant ( $f$ ), death rate relative to birth rate ( $d$ ), and plating efficiency/sampling fraction ( $p$ ). It supports piecewise-constant parameters, allowing for multi-phase growth dynamics (e.g., varying death rates over time).
- Validation: The simulator was validated against analytical methods (where applicable) and the standard Gillespie algorithm, demonstrating statistical indistinguishability from analytical results and a significant speedup over the Gillespie method.
Approximate Bayesian Computation (ABC-MCMC):
- Inference Engine: The simulator is coupled with an ABC-MCMC algorithm using a Metropolis-Hastings approach.
- Process: The algorithm explores the parameter space by generating simulated distributions of mutant counts. A new parameter set is accepted based on the resemblance between the simulated distribution and the observed experimental distribution, measured using the Kolmogorov-Smirnov two-sample test statistic.
- Likelihood-Free: This approach allows for parameter inference without requiring an explicit likelihood function, which is often intractable for complex demographic scenarios.

Key Contributions and Results
The study validates the proposed method across a broad range of biological scenarios and compares its performance against three state-of-the-art analytical tools: rSalvador, Flan, and mlemur.

Standard Conditions: In idealized scenarios (no death, equal fitness, full plating), the ABC-MCMC method yields estimation errors comparable to analytical tools, though with slightly higher variance due to stochasticity. Accuracy for all methods is primarily limited by the number of replicate populations and the median number of observed mutants.
Known Non-Standard Conditions: When parameters such as fitness ( $f \neq 1$ ), sampling ( $p < 1$ ), or constant death ( $d > 0$ ) are known and provided to the estimators, the ABC-MCMC method performs comparably to mlemur and rSalvador. rSalvador and Flan fail to provide accurate estimates when cell death is present.
Complex Simultaneous Conditions: The method demonstrates unique capability in scenarios combining fitness effects, cell mortality, and partial plating.
- Piecewise-Constant Death: The ABC-MCMC method successfully infers mutation rates in multi-phase growth models with varying death rates, a scenario where mlemur (limited to constant death) and other tools fail.
- Joint Inference: The method can simultaneously infer mutation rate ( $\mu$ ) and relative fitness ( $f$ ) with accuracy comparable to rSalvador and Flan in standard conditions. Crucially, it remains the only tool capable of jointly inferring $\mu$ and $f$ in the presence of known cell death and sampling.
Limitations in Identifiability: The study reveals that jointly inferring mutation rate ( $\mu$ ) and death rate ( $d$ ) when both are unknown is statistically unreliable. The authors demonstrate that $\mu$ and $d$ are not identifiable from mutant count distributions alone; their effects on the distribution are indistinguishable. Consequently, the method requires $d$ to be known (via external experiments) to accurately infer $\mu$ in the presence of mortality.

Significance
The paper claims that this simulation-based approach provides a flexible alternative to analytical models for estimating mutation rates in arbitrarily complex conditions. Its primary significance lies in its ability to:

Circumvent Restrictive Assumptions: It handles scenarios involving cell death, unknown fitness effects, and partial plating simultaneously, which are common in stress-response studies but incompatible with existing analytical tools.
Extend Modeling Capabilities: It allows for the inference of parameters in piecewise-constant growth dynamics (e.g., stress followed by recovery), a feature not supported by current software.
Conceptual Simplicity: By relying on forward simulation rather than complex analytical derivations, the method is conceptually straightforward and can be adapted to new biological hypotheses (e.g., mutation rate heterogeneity, phenotypic lag) as long as they can be integrated into the simulator.

The authors conclude that while analytical methods remain preferable for their speed in simple scenarios, the proposed ABC-MCMC framework is essential for accurate parameter inference in the complex ecological and physiological conditions frequently encountered in modern evolutionary microbiology.

More like this