A single fermented food sample can contain thousands of different microorganism species. Figuring out which ones are actually useful, the classical way, means isolating them, growing each one in culture, and then testing it. That process can take months, sometimes years. Now imagine skipping most of that — predicting the answer directly from genetic data, without ever stepping into a wet lab.
That's exactly where artificial intelligence is stepping in — and the pace here has genuinely picked up over the last year or two.
Discovery driven by data, not guesswork
The Korea Food Research Institute recently announced a library of 2,000 strains: 1,500 collected from the gut microbiome, 500 from fermented foods. The library itself isn't the interesting part — what matters is that the AI models running on top of it can predict which strain is likely to perform which function before anyone touches a petri dish. Instead of "let's culture this and see what happens," the logic becomes "this genetic profile points to that function, so let's prioritize testing it."
This isn't just an academic curiosity, either. Companies like Ingredion and Shiru are scanning over 77 million natural protein sequences on dedicated platforms to discover new functional ingredients. The logic is the same: rather than screening millions of possibilities by hand, you tell a model what properties to look for and let it search.
So how does this connect to what we do?
We already use metagenomic sequencing — reading the entire microbial community of a fermented product directly from its DNA, without trying to culture it first. Up to now, interpreting that data has relied heavily on our own experience, and that of similar labs. What AI adds is the ability to make predictions like "this strain likely produces bacteriocins" or "this one carries a biogenic-amine risk" — across thousands of candidate strains, much faster and at much greater scale.
There's a flip side, of course, and researchers in the field are upfront about it: data standardization is still weak, results from different labs don't always line up cleanly, and it's often unclear why a model made a given prediction. So AI isn't replacing wet-lab testing — not yet, anyway. What it does is tell you which strains deserve a first look, setting the priority order for the search.
What actually changes in the short term?
The biggest impact is probably going to be speed. Today, getting a starter culture combination from the lab to industrial scale takes years — we went through that ourselves with the BAKTOGARD® sucuk culture. AI-assisted strain prioritization could meaningfully shorten at least the first stage of that process: narrowing thousands of candidates down to the reasonable ones. What it looks set to replace isn't the wet lab — it's the trial and error.