Gradient Biotech
U

Menu

Research Areas

Computational BiologySingle-cell, bulk RNA-seq, spatial, and biomarkers
CardiologyCardiovascular signals, cohorts, and interpretationImmunologyImmune cell states, signaling networks, and repertoire dynamicsNeurologyBrain connectivity, functional imaging, and electrophysiologyOncologyTumor microenvironment, immune response, mutations, and survivalPathologyWhole-slide imaging, tissue quantification, and spatial pathology

Application

Documentation

Documentation

Back to studies

AI interpretation

Immunology interpretation follows the same AI pattern as Cardiology: grounded prompts, Ollama gpt-oss:120b, JSON output, and post-response guardrails.

What gets interpreted

The interpretation pipeline reads completed Immunology run records and extracts citeable metrics from:

  • immune composition
  • annotation and state scoring
  • repertoire analysis
  • cell-cell communication
  • disease workflows
  • multi-modal profiling

Interpretation and report runs are excluded from the source context so reports cite scientific analysis runs only.

Guardrails

The pipeline enforces:

  • every hypothesis cites computed outputs
  • cited run IDs, metric keys, and values must match context
  • fabricated numeric values are blocked
  • clinical language and treatment recommendations are blocked
  • research-only disclaimer is included

Model configuration

The app uses Ollama cloud when configured:

OLLAMA_API_KEY=...
OLLAMA_BASE_URL=https://ollama.com

Model: gpt-oss:120b

If the service is unavailable, a deterministic fallback interpretation is generated and marked in the audit artifact.