Skip to main content

From your separation targets to a predictive model

Tell Studio your chromatography mode and chemistry and upload your runs. Studio fits the candidate isotherms, ranks them on how well they reproduce your data and builds a process model. Simulate across the flow rates, gradients and buffers you could run, and you see which conditions reach your separation targets, and whether they can be reached at all, before you pack a column.

Book a demo

A model of your separation, from the runs you have

You bring the run data you already have and describe the column, the resin and the dissolved components. The chromatography model workflow in Novasign Studio turns that into a process model, for bind-and-elute and flow-through operation across the common chemistries, ion exchange, hydrophobic interaction, affinity and mixed-mode, and for a product from any modality. The models run on Novasign’s own solver, built for fitting isotherms and process parameters to your runs and for optimizing on them, so choosing the isotherm and the column model is not a decision you have to make up front.

With the model trained, you can try conditions that have never been run on a column and see how the process would have behaved. The model suggests which conditions are worth testing next and which flow rates, gradient lengths and buffer compositions move you towards your separation targets, and you decide which of them are worth a run.

You define the model once and retrain it on every new experiment, from early development across scale-up to manufacturing.

The supported operations

The workflow describes every operation a chromatography step is built from with the same handful of concepts: the column and its resin, the molecules in the load and the buffers that are added. What changes between a capture and a polishing step is which chemistry binds the product and which buffers drive the phases, and the way you describe the column stays the same.

Equilibration, load, wash and elution phases define the chromatography process, and every one of them can be defined in the model, optimized and used for what-if scenarios to improve the process.

Bind-and-elute capture

What it does
Binds the product to the resin while the load passes, then elutes it on a gradient or in a step
How it is described
A load, a wash buffer and an elution buffer, with the product in the eluate

Bind-and-elute polishing

What it does
Separates the product from closely related variants on the resin
How it is described
The same buffers, with the gradient length setting the resolution

Flow-through

What it does
Lets the product pass while impurities bind
How it is described
A load and a wash buffer, with the product in the flow-through

Each chemistry is described by its binding model, such as Langmuir or HIC Unified, or one identified from your runs, and the column by its transport model: the lumped rate model with pores, the general rate model or the equilibrium-dispersive model. Contact Novasign if your operation is not mentioned.

You describe the separation. Studio sets up the model.

Setup has two halves, your chromatography process and the machine learning.

In the words you already use

Each half keeps its own vocabulary: process parameters and quality attributes for your process, loss functions and checkpoints for the machine learning.

The bioprocess side: parameters, components and buffers

With the runs in, the setup walks you through your process.

Chromatography Model
1

Initial configuration

The column height and diameter, the resin the column is packed with and the characteristics that describe it, particle size and pore size among them, the time base, and the run IDs that separate one run from the next.

2

Process parameters

The parameters that are recorded and actively set during the run, and that the model should account for: flow rate, gradient length and phase length, among others, and how each phase was run, as a step or on a gradient.

3

Components

Everything that is in the system or can enter it. Set the product, product variants, impurities and buffer salts and define your material attributes and quality attributes, each with its own transport parameters, the diffusion coefficient among them, which depend on the pairing of resin and product rather than on either alone.

4

Buffers

Every buffer and flow into or out of the column that was recorded: the equilibration, wash and elution buffers and the load on the way in, the flow-through, the eluate and the waste on the way out.

Every data column is defined with its unit, cm/h, bar or g/L, so the mass balances stay dimensionally consistent and nobody has to rescale a column by hand to make them add up.

Any of them can be a condition of your experimental design, column size, flow rate, gradient length, buffer and load concentration and time among them, so a campaign that varied the column trains the same model as one that varied the gradient.

The machine learning side: model training, tuning and selection

Trials save checkpoints as they train, and the model you carry forward is selected from them, as for every Studio model.

  1. 1

    Train

    Training fits the model to your data, with defaults tuned for chromatography processes rather than for machine learning in general. Where chromatography gets a setting of its own is the isotherm: pick the candidates to fit from the library of adsorption isotherms, or fit them all, and where none of them should be assumed, a data-driven isotherm infers the binding from the run data alone.

  2. 2

    Tune

    Tick tunable next to a setting and the tuner sweeps it in one pass instead of one candidate at a time.

  3. 3

    Select

    The selector chooses between the candidates. For the isotherm it proposes the candidate with the lowest error and draws every fitted one over the data, automatic by default and yours to override when a worse score is the better isotherm. For the rest, it selects automatically or against criteria of your own.

Find out whether your targets can be reached

With a trained model you can simulate runs that were not performed. The digital twin takes the trained chromatography model, adds a simulation configuration, and returns time-resolved simulations of everything the model was trained on: the process parameters, the product and impurity concentrations, and the quality attributes, whether that is the chromatogram of a gradient you have not run or the gradient length at which two peaks come apart.

The decisions the digital twin helps you make

Which decision you face depends on the phase of the lifecycle the process is in.

  • Screening

    Which resin and process parameters influence the process most?

    Vary them across the space and watch what the breakthrough, the elution and the critical quality attributes do. A parameter that moves nothing is one fewer to characterize.

  • Optimization

    Which conditions give the best performance at acceptable quality?

    State the objective and the ranges to optimize within, and run the simulations inside that space. Every in-silico run can be read in a table or a chart. The visualizations and statistics show you which experiments are worth performing.

  • Characterization

    How robust is the process inside its design space, and over cycles?

    Restrict the acceptable ranges of your critical quality attributes and what remains is a design space. Vary the process inputs inside it, and how much the outputs move is your answer on robustness. A column parameter such as the porosity can follow the cycle count, so resin aging is part of the same question.

  • Scale-up

    Which conditions are suitable for the move to a larger column?

    The twin accepts any column size: set the geometry of the larger column and keep the resin and the components, because the isotherm belongs to that pairing and not to the column. Runs from columns of different sizes train one model the same way, so what gave acceptable results at lab scale can be checked against what the larger column allows.

Configuring a simulation

A simulation adds the initial state of the column and the phases of the process to what the trained model already knows.

The initial state is the buffer the column is sitting in when the run starts. Changing it is how you simulate a column arriving from a different hold or a different preceding step.

A phase is a section of the chromatography process, and which ones you have depends on the method: equilibration, bringing the column to the conditions the load needs; load, where the product binds to the resin or passes through it; wash, moving what is not wanted off the column; and elution, bringing the product back off on a gradient or in a step. Each phase is defined by three things.

1

Process parameters

Every phase has its own set of setpoints.

2

Buffers

The inflows are the load and the buffer that drives the phase. The outflows are the flow-through, the eluate or a sampling stream.

3

Stop criterion

A duration, a volume, a number of column volumes, or a component falling below a threshold. The first to fire ends the phase and starts the next.

Chromatography Process Simulator

Generating the in-silico runs

The digital twin lets you run the process in silico for a new set of inputs.

Chromatography Use Case

Simulation inputs

1x (0.75·10¹² cp/mL)
245 MV
99 %

Load titer, 4 mL feed 1x (0.75·10¹² cp/mL); Gradient length, MV (membrane volumes) 245 MV; Purity target of the pool 99 %; Full capsids recovered 63.5 %; Pool over load concentration 0.128×

Predicted value

  • Full capsids recovered

    63.5 %

  • Pool over load concentration

    0.128×

Training data from Ossi, Delfino, Marie, Schulte and Sponchioni, “Anion-Exchange Membrane Chromatography for the Separation of Empty and Full Adeno-Associated Viral Capsids”, Biotechnology Journal 2026, 21(8):e70295, doi.org/10.1002/biot.70295

In the ECONTI research consortium, a Novasign hybrid model timed the column switches of a continuous multi-column chromatography while the process ran, predicting the loading conditions from the measured product concentration rather than from a clock. Novasign predicted, and the automation platform decided and actuated.

Bring one resin and one run. Leave with a method you can set.

Bring the chromatography step you are developing and the resin you are developing it on. We design the first experiment with your team, run it, and build the model with you in Novasign Studio, so what comes out is a process model of your own chromatography step. From there the workflow is yours: design the next experiment, retrain on it and run the simulations, with us alongside for as long as needed, or on your own from the first run. Not a standard unit? Scientific modelling models it with you.

Book a demoGet in touch
Related

Downstream modelling, written up

A hybrid model timing a continuous purification, and the method and software behind it.

All posts