METIS

Ask a biological question.
Get an auditable strain shortlist.
Decision support for faster, safer strain discovery
BIOTECHNOLOGY + AI
24–25 September 2026 · Toulouse
Marco Anteghini (LifeGlimmer)
METIS
Bioindustry4.0

Strain discovery starts with one question

The data exist. The decision layer does not.
One biological question
Which safe strain can break down cereal fibre?
BacDive
UniProt
NCBI
MIRRI
StrainInfo
Literature
Internal files
Patents
The cost
TIME
lost to the wrong starting strain
+
hallucinated accessions
+
biosecurity risk
Metis connects the sources and keeps track.
METIS
Bioindustry4.0

Do not ask the model to know. Ask it to translate.

The LLM translates intent into query syntax. BacDive supplies the biological facts.
Human question

“Find strains isolated from soil in Germany.”

Generated SPARQL
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#> PREFIX d3o: <https://purl.dsmz.de/schema/> SELECT ?strain ?id ?source WHERE { ?strain a d3o:Strain ; d3o:hasBacDiveID ?id . ?isolation a d3o:IsolationSource ; d3o:describesStrain ?strain ; rdfs:label ?source . FILTER(CONTAINS(LCASE(STR(?source)), "soil")) ?origin a d3o:LocationOfOrigin ; d3o:describesStrain ?strain ; d3o:hasCountry "Germany" . } LIMIT 12
Live BacDive results
Myxococcus xanthusID 10011 · soil from a stream bank
Myxococcus xanthusID 10012 · soil
Myxococcus xanthusID 10035 · soil with straw
Myxococcus xanthusID 10040 · soil
12 records · first 4 shown
This simple division of labour was the starting point of METIS.
METIS
Bioindustry4.0

One controlled path from question to grounded answer

Natural language in. Ranked, source-linked evidence out.
Ask

ASK

Plain-language
question

Translate

TRANSLATE

SPARQL · REST
E-utilities

Check

CHECK

Risk filter
+ audit log

Retrieve

RETRIEVE

Trusted
databases

Ground

GROUND

Ranked shortlist
+ provenance

Databases stay the source of truth. Every shortlisted row carries field-level evidence; every output is safety-checked and auditable.
METIS
Bioindustry4.0

A real question became a verified strain shortlist

Brewer’s spent grain  •  risk-group-1 strains  •  cereal-fibre breakdown
Step 1BacDive7 risk-group-1 species that break down cereal fibre
Step 2UniProtenzyme evidence per species, e.g. Swiss-Prot P22222
Step 3Collections18 accessions for the 7 strains, from BacDive records
Step 4NCBIsupporting literature on spent-grain valorisation
METIS / workflow result
18 records  •  showing 5  •  BacDive, 10 Sep 2026
name cultureCollectionNumber collection riskGroup
Acetivibrio cellulolyticus ATCC 33288 ATCC 1
Acetivibrio cellulolyticus DSM 1870 DSMZ 1
Actinophytocola xanthii DSM 109075 DSMZ 1
Actinophytocola xanthii KCTC 39690 KCTC 1
Cellulosimicrobium cellulans CIP 104849 CIP 1
Every row carries its source. Nothing is invented.
METIS
Bioindustry4.0

We tested it where chatbots fail

14 obscure type strains, chosen so no model could have memorised their accessions
100%
Metis: correct culture-collection accession
36%
standard chatbot, same question, no database
+
74% of the chatbot’s claims were unsupported by any record.
+
Every disagreement went Metis’s way. McNemar p = 0.004.
+
Same 100% on three independent engines (DeepSeek · GPT-4o · Llama-3.3-70B).
One strain from the blind set
Aeoliella straminimaris: which collection has it?
Chatbot
DSM 100188 · JCM 30882
Confidence: “high”. Uncertainty: “none”. Complete with a citation.
Metis
CECT 30574 · DSM 114064
Retrieved from BacDive, field-level provenance, snapshot 2026-07-01.
Two invented accession numbers, delivered with full confidence. You only find out after you have ordered the strain and waited.
The database link does the work, not a bigger model.
METIS
Bioindustry4.0

From evidence to your next experiment

Working prototype today: four databases live, benchmarked, preprint out.
Here is what we are asking for.
Book a Metis working session
anteghini@lifeglimmer.com
marcoanteghini@gmail.com
Scan to book a session: Mondays and Fridays, October to November.
Team

Marco Anteghini  •  Lorenz Reimer

Vítor A. P. Martins dos Santos

and the whole BIOINDUSTRY 4.0 WP6 team

Next 12 months: connect more collections, run pilots with industrial partners, open the catalogue layer.

Culture collections & mBRCs

Structure your catalogue with us: more visibility, more strains actually used.

Industrial partners

Bring one real strain question. We run it with you as a pilot.

Data access

Endpoints and internal strain tables we can connect next.

The vision Metis is the layer between the two sides: weeks of strain search become minutes, projects start from better strains, and every connected collection makes the next answer more valuable, for industry and for the collections alike.