Skip to main content

Gene expression

Regression that predicts gene expression from DNA sequence together with an experimental context. Output is on a log(TPM+1) scale. The model is multi-modal: a ModernBERT DNA encoder combined with a Qwen3 text encoder and a decoder, trained on ENCODE RNA-seq data across cell types.

Input​

This task requires two inputs:

  • A DNA sequence in a fixed 9,198 bp window centered on the transcription start site (4,599 bp on each side of the TSS), passed as sequence or a server-side handle.
  • A mandatory options.description prompt describing the experimental context (for example the cell type). The text encoder consumes this prompt, so it is required for every prediction.

Output​

A single expression value on the log(TPM+1) scale, returned as json in the {data, meta} envelope.

Try it​

REST API: POST /v1/tasks/expression/predict with options.description set — see the getting-started guide. MCP: "Predict HBB expression in K562 cells." See mcp.md.