Microbiome-based therapeutics have accumulated a compelling body of clinical evidence over the past decade. What they have lacked is the manufacturing and design infrastructure to turn that rationale into reproducible, scalable drugs. Munich-based mbiomics is attempting to build that infrastructure.
The company's platform combines AI and machine learning-driven consortia design, proprietary analytical technology, and large-scale consortium co-cultivation capabilities — a technology stack it argues has been the missing ingredient in microbiome drug development. Its lead program, MBX-116, is a defined bacterial consortium being developed as a co-therapy to immune checkpoint inhibitors (ICIs) in second-line advanced melanoma.
Why microbiome manufacturing is challenging
The core engineering challenge is one of biological complexity. A therapeutic microbiome consortium is not a single organism optimized for production — it is a community of organisms, in mbiomics' case around 150 bacterial strains, each with its own fermentation window, oxygen tolerance, and nutrient requirements, that must arrive at the patient at defined ratios and defined viability.
"Every strain has its own fermentation window, and co-cultivation forces you to balance competition, cross-feeding, and ecology rather than optimize a single organism," said Johannes Woehrstein, cofounder and CEO of mbiomics, told DDN. "What held the field back is the missing technology stack: Computational models to design the communities, and an analytical layer that resolves consortium composition at strain level."
That strain-level resolution during the fermentation process — not just in the finished product — is what mbiomics says distinguishes its analytical approach. Tracking each strain's abundance and viability trajectory through co-cultivation allows the company to identify when a member is being outcompeted, when cross-feeding is failing, or when viability is dropping, and to use that information to refine the process toward tighter control and more reproducible community composition.
AI at two levels
The computational layer of the platform operates at both the strain and consortium level. At the strain level, models predict each organism's functional profile — its metabolic and immunological output — and how well it fits into a given community context. At the consortium level, the models predict community behavior: which combinations are ecologically stable, which produce the desired metabolic and immunological functions, and which are likely to engraft in the relevant patient population.
"A strong computational platform of this kind lets us design candidates in silico, test them at scale, and feed the results back into the next design cycle," Woehrstein said. "The practical effect is that strain selection stops being the bottleneck. The bottleneck moves to how fast we can build, characterize, and screen the designed communities."
That iterative design-build-test cycle is central to how mbiomics moves from a mechanistic hypothesis to a clinical candidate — each stage feeding results back into the design layer for the next round of refinement.
The melanoma rationale
The choice of second-line advanced melanoma as a lead indication reflects a deliberate bet on one of the better-evidenced microbiome-disease links in oncology. Studies examining melanoma patients undergoing anti-PD-1 (programmed cell death protein 1) immunotherapy have found significant differences in the diversity and composition of the gut microbiome between responders and non-responders. Clinical trials have demonstrated that fecal microbiota transplants from melanoma patients who responded to anti-PD-1 therapy could overcome resistance in roughly a third of patients who had progressed on treatment.
That said, the field has also grappled with the complexity of the microbiome-ICI relationship. Cross-cohort analyses have found that while the gut microbiome shows a relevant association with ICI response, the reproducibility of specific microbiome signatures across patient populations is limited, and no single species has emerged as a fully consistent biomarker. That heterogeneity is part of what mbiomics argues makes a defined, engineered consortium — rather than a donor-derived fecal microbiota transplant — a more tractable therapeutic approach: It allows mechanistic targeting of the immunological functions believed to drive response, rather than transplanting a complex and unpredictable donor community.
The company runs extensive in vitro screening and characterization of consortia to measure function, metabolite production, and interaction with host cells, alongside in vivo models that demonstrate anti-tumor activity and synergy with ICIs. Beyond oncology, mbiomics is building a broader pipeline across autoimmune and neurodegenerative disease indications, where the same design and screening platform applies but the functional readouts differ.
