To enjoy the full framework and story, see our Perspective “Modelling the role of the microbiome in antimicrobial resistance across scales” in Nature Microbiology.

The spark at the Collegium

Two days in Zürich, a room full of people who study antimicrobial resistance (AMR) in very different ways, and a question that kept coming back: are we talking about the same thing? That workshop, organised by Lisa Pagani, with Sonja Lehtinen, Roger Kouyos and Sebastian Bonhoeffer at the Collegium Helveticum, is where this Perspective began.

The discussion groups after the seminars moved quickly across scales, from ecological interactions within microbial communities to treatment effects and transmission between hosts. The notes and outline from those discussions became the starting point for shaping the manuscript.

Even though we deal with the same problems and biological questions, we are not always speaking the same language. At the bench, many processes remain entangled, as they happen simultaneously and cannot be fully separated. Growth, competition, gene transfer, antibiotic exposure and community structure often unfold together. In a model, we have to separate them. We have to decide what to keep, what to leave out and what biological aspect the model is really trying to represent. And those decisions are usually not neutral.

Finding a common language between the bench and the models became one of the central aims of the paper.

Poster for the Collegium Helveticum workshop “Quantifying the Impact of Microbiome Dynamics on Antimicrobial Resistance” in Zürich, where this Perspective started.

From isolated pathogens to communities

Resistant bacteria do not exist in isolation. They emerge, compete and persist within complex microbial communities. For example, plasmids carrying antibiotic resistance genes can spread across different species within the gut microbiota of hospitalised patients, and from patient to patient. Such cases highlight the clinical relevance of microbial-community-level interactions.

The challenge is to understand when these interactions matter. Sometimes, the microbiome may determine whether a resistant strain can invade. Sometimes it may act as a reservoir of mobile resistance genes. Sometimes, antibiotic treatment may reshape the community in ways that create new ecological opportunities for resistance to persist or spread.

It is tempting to respond to this complexity by adding more biological detail to mathematical models; more species, more genes, more compartments. But greater biological realism does not necessarily produce a better understanding. The question is not how to include everything. The question is what a model needs to represent to explain the process we care about.

What should the model see?

Microbiome-aware AMR models should be built around specific biological questions. A model designed to study antibiotic disruption of colonisation resistance will not need the same structure as one focused on plasmid persistence, spatial organisation in biofilms or transmission between human hosts. In each of those cases, the relevant scales change.

This is where modelling becomes a way to decide which part of the biology is carrying the causal signal. In some cases, the key process may be ecological competition. In others, it may be stochastic loss, spatial structure or movement between hosts. We often face these decisions in our own projects, and sometimes the most useful model is the one that deliberately leaves things out.

This also changes how we think about data. Microbiome-AMR models often need information from very different sources, from controlled experiments to metagenomic time series and epidemiological studies. No single dataset captures the full process. Models can help connect these partial views, but they can also reveal which measurements would be most informative to make next.

In the end, modelling AMR in the microbiome is not about making existing models more complicated. It is about changing what the model is allowed to see. The microbiome is not a passive scenery around resistant bacteria. It is an ecological system that can shape the fate of resistance. The challenge is to identify which parts of that system matter and how to represent them without losing the biological question, and that is exactly what models (when done well) are for.

A connected view of antimicrobial resistance across scales. This illustration represents the microbiome as an interconnected ecological system linking genes, microbial communities, hosts and environments. Antimicrobial resistance can emerge, persist, and spread across these interconnected scales, while mathematical models provide a way to track, interpret, and predict these trajectories. The image is original vector artwork, drawn manually in Inkscape.