Skip to content
Cerebra BioSystems

Architecture

How the pieces connect.

Every component runs on your network. Hover over or tab to a step to see what it does.

Your lab network

  1. The pipeline computes

    Your instrument writes run data to shared storage, just as it does today.

  2. Network storage in your lab holds raw data and results, with scheduled snapshots.

  3. Detects when a run is fully written, checks the sample sheet, and starts the right pipeline.

  4. Nextflow and nf-core pipelines in containers, with pinned versions so results are reproducible.

  5. Every run gets a record of its inputs, pipeline version, settings and checksums.

  6. The AI explains

    Read-only tools that let the AI look up QC, expression results and logs, within each user's permissions.

  7. An open-weight language model running on your AI computer. It never connects to the internet.

  8. A browser-based chat your scientists sign into with their company account.

Text description of the diagram

A run moves through eight components, in order. The first five process the data; the last three let your team ask questions about the results.

  1. Sequencer (run folder): Your instrument writes run data to shared storage, just as it does today.
  2. Lab storage (NAS · snapshots): Network storage in your lab holds raw data and results, with scheduled snapshots.
  3. Run watcher (sample sheet check): Detects when a run is fully written, checks the sample sheet, and starts the right pipeline.
  4. Pipelines (Nextflow · nf-core): Nextflow and nf-core pipelines in containers, with pinned versions so results are reproducible.
  5. Results + run record (run_manifest.json): Every run gets a record of its inputs, pipeline version, settings and checksums.
  6. Data tools (MCP · read-only): Read-only tools that let the AI look up QC, expression results and logs, within each user's permissions.
  7. Local AI model (open-weight LLM): An open-weight language model running on your AI computer. It never connects to the internet.
  8. Private AI workspace (browser · SSO): A browser-based chat your scientists sign into with their company account.

Step by step

What happens at each step.

  1. 01

    Run detection

    The run watcher waits until the sequencer has finished writing and the files have stopped changing, then validates the sample sheet. Problems, like a missing sample or a mismatched barcode, are flagged before any compute starts.

    What your team sees

    The run is picked up on its own. Any sample sheet problem is reported before processing begins.

  2. 02

    Pipelines

    We configure the nf-core pipelines your assays need, with your reference genomes and QC thresholds. Pipelines run in containers with fixed versions, and each run is recorded so it can be re-run later with identical results.

    What your team sees

    A status for every run, from queued to finished, with nothing to launch by hand.

  3. 03

    Analysis and reports

    Beyond raw processing, we set up the downstream analysis your team actually uses: differential expression, pathway enrichment and standard figures. A report is generated automatically for every run.

    What your team sees

    A report for each run with QC flags, differential expression tables and figures.

  4. 04

    Private AI

    The AI reads results through read-only tools. It explains and summarizes; the statistics come from the pipeline.

    What your team sees

    Plain-English answers, with every number linked to the file and row it came from.

Supported assays

Supported assays and pipelines.

Supported assays, the pipeline used for each, what you get and status
Bulk RNA-seqPipelinenf-core/rnaseq+ nf-core/differentialabundanceWhat you getQC report, gene counts, differential expression, pathway results, figuresStatusAvailable
Single-cell RNA-seqPipelinenf-core/scrnaseqWhat you getQC, count matrices, clustering-ready dataStatusOn request
WGS / WESPipelinenf-core/sarek (GPU-accelerated options)What you getAligned reads, variant calls, QCStatusOn request
ATAC-seqPipelinenf-core/atacseqWhat you getPeaks, QC, coverage tracksStatusOn request
Amplicon / 16SPipelinenf-core/ampliseqWhat you getTaxonomic profiles, diversity, QCStatusOn request
Something elsePipelineCustom Nextflow pipelineWhat you getScoped with youStatusAsk us

On request means we scope and validate the pipeline for your lab, usually during the pilot.

Reproducibility

Every result can be traced and re-run.

Each run produces a run record listing its input files and checksums, the pipeline and version, the container images, the reference genome and every setting. Re-run it in a year and you'll get the same answer.

  • Pinned pipeline versions
  • Container images pinned by digest
  • Input files recorded with checksums
  • Re-runnable later with identical results
run_manifest.json
{
  "run_id": "run_0142",
  "pipeline": "nf-core/rnaseq",
  "pipeline_version": "3.x.x",
  "reference": "GRCh38 · GENCODE v4x",
  "samples": 24,
  "inputs_sha256": "…",
  "containers": "pinned by digest",
  "completed_at": "2026-11-12T03:41:00Z"
}
Example run record (abbreviated)

The AI's role

Clear limits, by design.

Does

  • Explain QC results and flag problems
  • Look up expression and statistics from your results
  • Diagnose failed pipeline runs in plain English
  • Search your papers, protocols and notes
  • Draft run summaries

Doesn't

  • Run its own statistics
  • Change, delete or move data
  • Answer when the data isn't there
  • Make diagnostic or clinical calls

Find out what your lab could automate.

30 minutes, no cost. You get a written recommendation afterwards.

30 minutes · No cost · No obligation · We'll send a written recommendation afterwards.