Praescigenics · Genetics industry intelligence

Catch the failing run before it fails.

A DNA sequencer is a biosensor at scale. A sequencing run is a monitored batch. The same foreknowledge that Praescia brings to biopharma manufacturing — applied to sequencing-run QC and regulated genomics manufacturing. Insight-only. Validation-ready. Powered by Atlas.

Request a demo Compare with Praescia

The problem

Your sequencing lab runs Illumina, Oxford Nanopore, or PacBio instruments producing millions of data points per run — Q-scores, cluster densities, error rates, reagent lot states. And a lab technician reviewing the run summary after it completes, writing up a failed-run report for a run that started showing drift 80% of the way through.

The signal was there. The detection wasn’t.

For regulated genomics manufacturing — viral vector production, plasmid manufacturing, cell and gene therapy — the stakes are higher still: a failed batch is not just a failed run, it is a failed deviation report, a potential patient impact, and a COGS event that could have been prevented.

What Praescigenics does

Praescigenics treats a sequencing run the way Praescia treats a bioreactor batch: as a monitored process with a golden trajectory, a live state, and an onset-of-deviation signal that precedes the visible outcome.

It reads real-time QC telemetry from the instrument — Q-scores, cluster density (Illumina) or pore occupancy (nanopore), error rates, reagent and flowcell lot state — fits a golden reference from good in-spec runs, residualizes each live run against it, and surfaces the excursion onset while the run is still running and intervention is still possible.

For regulated genomics manufacturing (vector, plasmid, cell/gene therapy), the same Praescia manufacturing spine applies directly — vendored unchanged, specialized only at the genetics leaf.

The architecture insight

A sequencer is a biosensor. A run is a batch.

Nanopore sequencing is ionic-current sensing at scale. Sequencing-by-synthesis is optical sensing at scale. Both produce a multi-signal process telemetry stream over time — exactly the shape the Praescia batch-intelligence spine was built for. Praescigenics is the genetics specialization of that proven machinery, not a new build from scratch.

What it catches

Mid-run

Quality excursion onset

%≥Q30 trending below the golden reference mid-cycle — surfaced before the run completes and the report is written. Time to decide: abort, yield partial data, or remediate loading.

On load

Cluster density / pore occupancy anomaly

Underloading (low yield) and overclustering (quality ceiling) detected at the loading checkpoint, not at the run summary. Illumina cluster density K/mm² · ONT pore occupancy fraction · PacBio P1 productivity.

Cross-run

Reagent & flowcell lot excursion

A bad reagent or flowcell lot appears across every run using it. Praescigenics surfaces the lot-level signal at the instrument level — before every run on that lot has failed. At density, this becomes a cross-lab early warning.

Error rate

PhiX / control error drift

Error rate departing the golden reference — the earliest quality signal before Q-scores or density reflect the problem.

Manufacturing

Vector / plasmid / cell-therapy batch deviation

Regulated genomics manufacturing batches monitored on the Praescia manufacturing spine — same golden-batch deviation detection, same GxP compliance shell.

Fleet (density tier)

Cross-lab lot-level signals

Federated QC telemetry (never raw reads, never patient data) surfaces reagent or instrument lot issues across labs simultaneously — the density signal no single lab could see alone.

The seven-stage genetics spine

The same seven-stage pipeline as Praescia — re-skinned for sequencing-run vocabulary.

Stage 1

OBSERVE

Ingest InterOp metrics (Illumina), sequencing_summary.txt (ONT), or SMRT data (PacBio). One adapter per platform. Not a fork.

Stage 2

NORMALIZE

Unify per-cycle / per-bin QC signals. Validate data integrity (ALCOA+). Align to the run’s time index.

Stage 3

DETECT_STATE

Map the run onto the inherited BatchStatus ladder: OnGoldenTrajectory → Recovering → Drifting → Constrained → Deviation.

Stage 4

DETECT_TRANSITION

Detect the onset of quality excursion or loading anomaly. Debounce against noise. Surface the transition early.

Stage 5

ESTIMATE_OUTCOME

Project run OOS-risk band (Low / Elevated / High / Critical) and estimated yield — called mid-run, not at the summary report.

Stage 6

RECOMMEND

One explainable advisory per event: “Review run — Q-scores trending off reference” / “Assess loading and reagent margin”. Named drivers, named disposition path.

Stage 7

VALIDATE

Deterministic replay, precision/recall tracking (when labeled data arrives), audit-trail generation. Every advisory replayable.

Platform support

Each platform is a registered profile — a reference envelope and instrument token set. Adding a platform is adding a profile; never editing the classifier.

Platform	Vendor	Loading metric	Status
NovaSeq / NextSeq / MiSeq / HiSeq	Illumina	Cluster density K/mm²	Synthetic rig end-to-end; InterOp adapter ready
Oxford Nanopore (PromethION, MinION)	Oxford Nanopore Technologies	Pore occupancy (fraction)	Gate-1a proven on real ONT telemetry
Revio / Sequel II	PacBio	On-plate P1 fraction	Profile defined; real-data gate open
Element AVITI	Element Biosciences	Cluster density (analog)	Roadmap
Manufacturing instruments	Platform-agnostic (via Praescia leaf)	Process telemetry	Inherited from Praescia manufacturing spine

The density architecture — super-linear at scale

Praescigenics has a stronger density argument than most infrastructure software. Sequencers of the same model share failure chemistry; reagent and flowcell lots are identical across every lab that purchases them. Failure signatures are highly portable. Federation converges fast.

L0 — Linear

Per-run monitoring

Each run is a monitored batch. The floor every QC competitor lives on. Priced $/run or $/instrument-month.

L1 — Margin expansion

Plug-in platform

New instruments, assays, and modalities snap on as plugins — not forks. Capability rises; marginal cost per run falls.

L2 — Super-linear

Lab / portfolio intelligence

Cross-instrument drift, reagent-lot tracking across runs, site scorecards. Sells to a higher-tier buyer (lab director, CLIA director, QA head) at higher price/run.

L3 — The moat

Federated intelligence

Failure signatures learned across all labs (anonymized QC telemetry, never reads, never genomes). A bad lot lights up simultaneously across every lab using it. A new entrant cannot replicate this at 3 instruments.

Federate telemetry, not genomes. The federatable signal is QC process metrics — Q-scores, densities, error rates. Raw genomic reads never leave the lab. This sidesteps HIPAA / CLIA patient-data liability and makes the L3 moat buildable under regulatory constraints. Enforced architecturally, not by policy.

Regulatory posture

Manufacturing context

Regulated genomics manufacturing (vector, plasmid, cell/gene therapy) inherits the full Praescia GxP compliance shell: 21 CFR Part 11 / EU Annex 11 / GAMP 5 / ALCOA+. Validation-ready by construction.

Clinical lab context

Sequencing-lab deployments operate in a CLIA / CAP context. Praescigenics stays decision-support throughout — advising the lab director and CLIA director, never producing a diagnostic result or becoming a regulated device (FDA SaMD posture maintained).

Insight-only is permanent. Praescigenics observes, predicts, and advises. It never actuates on a sequencing instrument, a manufacturing process, or a run in progress. The regulatory moat is architectural, not a configuration switch.

Evidence status (honest scope)

Gate	Status	What it proves
Gate-0 — synthetic rig	Complete	Plumbing end-to-end on a synthetic sequencing-QC rig. Does not credit detection.
Gate-1a — real unlabeled ONT telemetry	Proven	Golden trajectory learned from real Oxford Nanopore runs (pycoQC). Departures rank correctly. No labeled ground truth → does not credit precision/recall.
Gate-1a — Illumina InterOp	In progress	InterOp adapter complete; golden-fit harness ready. Awaiting real production InterOp run set.
Gate-1b — scored detection (labeled data)	Open — needs design partner	Real runs + lab disposition labels (passed / reworked / failed / OOS). Precision, recall, and lead-time credited here. This is the design-partner deliverable.

Design-partner program — Praescigenics

We are looking for one sequencing lab or genomics manufacturer with real run data — Illumina, ONT, or PacBio — to run a fixed-fee pilot. Your runs prove the detection; you get the findings and the first-reference status.

Apply for the program Compare with Praescia