Chromatography control
Load, gradient, and pooling set live from in-line UV, conductivity, and pH.
Downstream control · purification
Recover more drug from every column.
Autonomous downstream purification that sets load, gradients, and pooling from live signals and predicts chromatography breakthrough before it happens.
Runs on Jetson Orin/Thor at the edge, DGX/HGX for training
The problem
Downstream quietly bleeds the margin upstream earned.
After the culture is harvested, purification decides how much of the expensive drug substance survives. Clarification, Protein A and ion-exchange chromatography, ultrafiltration and diafiltration, and viral steps have to strip everything but the target to parts-per-million purity without denaturing it, over Protein A resin that costs thousands per liter and membranes that foul.
Load, gradient, pooling, and cycle decisions are still made from fixed recipes and offline assays. A mistimed pool, an early breakthrough, or a missed viral step loses yield, wastes resin, or scraps a batch worth millions.
What Purion does
Closed-loop control across the whole downstream train.
Purion takes control of clarification, Protein A and ion-exchange chromatography, UF/DF, and viral steps. It sets load, gradients, and pooling from live in-line signals and Titeon's CQA readings, and predicts breakthrough before it happens. It optimizes resin cycles and buffer prep to squeeze yield and purity out of every run while holding the purity and viral-clearance spec. Every action is bounded and logged to a validated audit trail.
Breakthrough prediction
Sees the breakthrough coming and adapts pooling to protect yield.
UF/DF and viral steps
Controlled with purity and clearance assurance built in.
Consumable savings
Resin-cycle and buffer-prep optimization to cut the cost per run.
Fits your stack
Connectors to ÄKTA and skids, DCS, and MES, with a Part 11 audit trail.
Learns across sites
Every run compounds into a shared purification-yield dataset.
- Yield and breakthrough prediction from in-line UV, conductivity, and pH
- Adaptive pooling and gradient control policies tuned per run
- Hybrid mechanistic and ML column models
- CQA-aware pooling grounded in Titeon's readings
Under the hood
Deciding the run, not just executing a recipe.
ÄKTA and skid controllers run fixed methods. Purion decides load, gradient, and pooling adaptively from live CQAs and learns yield across sites. It closes the loop where the incumbents only execute, and the adaptive policies sharpen as they train on partner runs.
Built on NVIDIA
GPU-essential during a running column.
- Edge hardwareJetson Orin and Thor (IGX) run skid decisions at the edge, where pooling cannot wait for a remote call.
- TrainingDGX and HGX train column and yield models over purification-run histories.
- LibrariesCUDA and TensorRT accelerate in-line breakthrough and pooling inference. cuOpt sequences chromatography, resin cycles, and buffer prep under constraints. RAPIDS accelerates the run-history feature pipelines.
- ServingTriton and NIM serve yield, breakthrough, and pooling models on site GPUs under NVIDIA AI Enterprise.
Commercials
Fast, provable ROI on yield and consumables.
- PricingBioseon Line module at $13,000 per purification skid per month, bundled into the Site tier, with outcome pricing on yield and resin-cost savings for enterprise.
- Who buysMSAT and downstream directors hold the budget, downstream process engineers champion it, QA approves.
- The moatA compounding cross-site dataset of chromatography runs, breakthrough curves, and pooling-versus-yield outcomes across resins and molecules.
- First deploymentIngest one chromatography skid, ship breakthrough prediction and advisory pooling, then move to bounded closed-loop control under human approval and add resin-cycle optimization.
Where it fits
Purion owns downstream, fed by the modules upstream.
Put your costliest suite on Purion.
A few points of yield or resin reuse pay for the platform. We land on the highest-cost purification suite first, then expand from there.