Pheno
How it works

From raw data to an organism your LLM can reason over.

Datasets in. A memory that grows. Answers with receipts.

Coding agents became useful the day someone gave them a memory of the codebase. Pheno is that memory for an organism.

Step one

Start with the data you already have.

Genome, proteome, expression, metabolites, papers.

Hand Pheno the genome, the RNA-seq runs, the metabolomics and the PDFs your group has collected. Each dataset is versioned on the way in, so every later answer can name the data it ran on.

FASTAv1GFFv2RNA-seqv3metabolomicsv4papersv5mediav6
What it takes in
Genome
FASTA assembly with GFF or GTF annotation
Proteome
protein FASTA
RNA-seq
differential expression per substrate and condition
Metabolomics
per-sample intensities and time courses
Literature
PDFs or identifiers
Media
named seed sets, or a list you supply
Step two

Agents turn it into knowledge that keeps growing.

Claims with sources, reviewed by you, compiled into releases.

Agents read each paper whole and propose claims with the quote and page behind them. You accept or reject, and what you accept compiles into the next release of the graph.

literaturereviewthe graphrelease 1release 2release 3
Step three

Ask your LLM. Get the answer with its evidence.

Routes, edits, structures and reports, served to Claude, ChatGPT or Grok.

Ask your LLM a strain question and it works through Pheno. The route, the drains and the structure come back with their evidence, saved to a report your lab can open.

the memoryartifactyour LLM
The idea

Memory layers, for biology.

Memory layers are what made coding agents useful on real codebases. Pheno applies the same idea to metabolism: your genome and your data are what it works on, the literature is the history behind them, and the graph is what your LLM remembers between questions.

Biology adds one step. A coding agent can run the code and find out whether it was right. Nobody can run a claim about an enzyme, so a scientist reads it against its source before it compiles.