The most frustrating thing for an R&D team is that, still today, microbial systems are, in a way, a black box. A probiotic strain looks potent in trials, but because the mechanisms behind microbiota dysbiosis are poorly understood, the same strain is given to patients with very different profiles, and efficacy is often poor. A biocontrol agent performs well in one field and disappoints in another. A fermentation that runs cleanly at bench scale starts drifting from batch to batch in a larger vessel.
Projects lose momentum, budgets come under pressure, and funding is eventually redirected to other projects with a clearer risk profile and a higher perceived probability of success, because teams can’t account for what exactly went wrong, where and why.
In a nutshell, today R&D teams can measure what goes into a microbial system. They see what comes out of it. But they often don’t understand what happens during the process, or why, and struggle to predict what conditions and strains are likely to perform according to plan.
A single strain carries a dense metabolic network and a community adds interactions that no single strain reveals on its own. Sequencing has become very affordable in recent years, so data is no longer a bottleneck. But capturing the “why” and “how” in data is a different story.
From screening to rational design: the first microbial digital twin
An obvious answer to complexity has always been extensive screening to test all potential hypotheses: test many strains, media, conditions, keep what works, and iterate. This approach is, of course, still at the very foundation of our innovation journey. But new capabilities to integrate and mine huge volumes of information and data are changing this paradigm: we can now switch from empirical to knowledge-based hypothesis formulation to rationalize our development processes.
We launched MINE, Microbial INtelligence Explorer, in September 2026, but it has been in the making internally for almost 10 years and it is, ultimately, the way our team decided to bring back to the industry over a decade of experience in microbial systems and R&D.
It works as a digital twin of the microbial system, guiding experimental design before work at the bench as well as supporting data interpretation and biological understanding afterwards.
It can rank candidate strains by their metabolic capacity, point to precursors and physicochemical parameters most likely to impact the production of a given molecule, flag where a process might behave differently at scale, and reveal which organisms are doing what in a system.
Design, experiment, decipher, learn: from complexity to clarity
In practice, a project with MINE can apply to the whole R&D cycle.
During the experiment design phase, MINE helps by selecting the strains with the potential to deliver the target function, identifying the parameters and precursors the experiment needs to test, and checking feasibility in silico. The solution can even model the potency of microbial combinations, and anticipate which strain could be boosted by which prebiotic. In summary, the solution enables the identification of the hypotheses worth testing at the bench.
During the experiment’s running phase, MINE Omics Project Management helps track all the experimental data: tested or patient parameters, samples being taken, replicates and more. Then (meta)genomic, (meta)transcriptomic and metabolomic data can be imported into the solution following robust data management principles: checks on data consistency, quality and assignment. This project management and data storage framework enables seamless collaboration between lab, clinical, analytics, sequencing, bioinformatics and biostatistics teams and offers organizations a unique platform to mine their microbial data.
Once all the experimental (omics) data has been gathered, MINE runs automated pipelines to analyze genomic, transcriptomic and metabolomic data to decipher what metabolic functions are present and active in the systems according to environmental and experimental conditions. This functional information is then integrated into a microbial knowledge graph database, which reconciles the multi-omics data. All data is fully quality-controlled and presented through intuitive data visualizations (multi-omics metabolic map, genome browser, evolution plots for molecule production, dashboards of metabolic capacities of microbial systems).
We learn the key experiment drivers through statistical analysis, mapping environmental/experimental data against omics results. Based on complex phenotypic features, our statistical and machine learning approach helps link observations to the phenotype behind them and explain what was observed. This is what makes the next design better than the last.
MINE starts from R&D teams’ needs and available data points
A common assumption is that a platform like this needs a full multi-omics dataset before it becomes useful or that it will force teams to work with entirely new strains. It does not, and that matters most early in a program, when data is thinnest.
Take an illustrative project: producing a targeted postbiotic compound by fermentation, from strain selection through to scale-up. With only the genomes of the strains as input, which could even be extracted from public databases, and by exploring the knowledge graph database to find the cascades of reactions leading to the postbiotic production, the solution found which precursors to feed, which physicochemical parameters the microorganism will respond to and which strains are the best candidates to produce the postbiotic because they carry the reactions it needs. When tested at the bench, our client found that the strain we predicted could produce the product, and the conditions we provided gave more than a twofold increase on the first try.
With genomic data on a candidate strain, we can directly identify the composition of the minimal medium required to grow it. We can also predict which carbon source, fiber, HMO or prebiotic the strain can grow on. Those elements are key to rationalizing all tests during microbial process development.
Using transcriptomic data collected during fermentation, we can track down the origin of batch-to-batch inconsistency, so corrective actions target the differences highlighted by MINE instead of proceeding by trial and error.
Making top-notch omics analysis available to non-omics experts
MINE is built for omics projects to be carried out across departments by non-omics experts. It is made of simple interfaces, user assistants and foolproof funnels so that R&D teams can generate and explore the data by themselves, without needing a bioinformatician or a statistician sitting next to them: comparable metabolic map extracts focused on the problem of interest, presented not in a computational way but in a way biologists can grasp; metabolic capability dashboards that map out which organism can do what; ready-to-use quality-control and statistical results reports; and an integrated genome browser for users who want to dig into genetics.
Running your omics project through MINE gives your R&D team the capability to explore the data, while relieving your bioinformatics and biostatistics team of routine, first-line analyses, from chaining open-source tools to searching text files, so they can focus on what brings value to your project.
People matter as much as software
Interpreting microbial data requires a skill set that not every customer has in full. Sometimes, we need a different approach.
For that reason MINE is available in two ways. Some partners want us to guide their project, with our team of biologists, biostatisticians and bioinformaticians running the analyses and returning interpreted results and qualified hypotheses to test. Others want the platform in their own team’s hands, with training and support. Partner data is hosted on-premise in an EU cloud, with top-notch data security and the highest compliance standards.
MINE has a longer history than its launch date suggests. Its foundations were laid through Microbiome Studio, an ML-enabled analytical platform born out of the Metapath project, an industrial program supported by the French government, along with Bel, Lesaffre and MetaToul’s top-notch metabolomics platform. We refined it on our own programs and through industrial collaborations before opening it to partners.
