AI agent for microbiologists
Metagenome Classification Review Agent
Produce a taxonomy table where disputed calls have been tested against an independent reference.
What it does
Different classifiers and databases can give different names for the same reads, and the choice often depends on which tool a person happened to run. This agent runs or reads the output from several taxonomy tools and databases on the same sample and compares them at each rank. It lists agreements and disagreements, and for the important disagreements it tests the contested calls against another reference, for example by aligning the reads to candidate genomes or checking marker genes. It flags calls that rest on very few reads or on a reference known to be incomplete. It recomputes the table with the tested calls and notes the confidence. It never decides the final taxonomy. The scientist approves the final calls. Edge case: a low abundance taxon found by only one tool is marked possible false positive.
How it works
Follow the arrows from top to bottom. The orange dashed arrow is the loop: when a check fails, the agent goes back and tries again.
Read the steps as a list
- Classifier outputs ready
- Load outputs from each tool and database
- Align calls across tools at each rank
- List agreements and disagreements above the abundance threshold
- Test each disputed call by aligning reads to candidate references
- Does the independent test settle each dispute?If not: try marker genes or a second reference and mark remaining calls unconfirmed. Back to step 5.
- Rebuild the taxonomy table with confidence notes
- Do the rebuilt totals match the read counts?If not: find the reads dropped or double counted and redo the table. Back to step 3.
- Scientist approves the final callsThe agent waits here for your OK.
- Reviewed taxonomy table
How it decides
It accepts a call when at least two tools agree or an independent test supports it. A call from one tool with few reads is held as unconfirmed.
- Accept a call when 2 or more tools agree at that rank
- Test any disagreement affecting a taxon above 1% abundance
- Hold a call as unconfirmed when it has under 50 supporting reads
- Record tool and database versions with every table
Make it yours
Every agent is a starting point. You choose these settings for your own situation.
- Tools and databases
- Abundance threshold for review (default 1%)
- Minimum read support (default 50)
- Reference sources
- Table format
What keeps you in control
It always asks you first
- The final taxonomy calls
- Any change in the abundance cutoff
Hard limits
- Never discard disputed calls, mark them
- Record all tool versions
It stops when
- Done: all major disputes settled or marked unconfirmed
- Stop: references too incomplete to test
Set it up
We guide you through the set-up, step by step
Members get the full set-up guide for this agent. No technical skills needed: you copy, paste and upload.
- One set of instructions to paste into your AI, with the clicks for ChatGPT, Claude, Microsoft 365 Copilot, Gemini and Grok
- The agent then walks you through connecting your own data, one source at a time
- A downloadable copy with the flow chart, the rules and the full guide
