Protein purification with AI — chromatography curve drift shows up before the peak, not after the batch.

The problem

Column drift shows up in the purity assay. By then, it is a batch of protein.

Therapeutic protein downstream (mAbs, recombinant proteins, ATMPs, vaccines) is a chain of steps — capture chromatography, ion exchange, mixed mode, TFF — where each step depends on the one before it. A subtle drift in the first column is amplified in the next one. And the final purity assay arrives when the batch is already made.

The system raises the alarm when the TFF pressure moves out of band or when peak yield drops. By then: the column has drifted, the buffer has been consumed, the purification cycle is compromised. The operator knew something was off — the curve was pulling away from the last good batch — but had no system that compared curves live.

The three pieces of data you would need to cross-check to anticipate the drift live on three islands:

1. The good batch signature — the UV, conductivity, pH and pressure curve of a correct batch on that same resin at that same cycle of use. In the historian, not tagged as a “reference signature”.
2. Today's real curve — the CPPs in the chromatography system or the TFF skid, second by second. In the equipment's SCADA, never cross-checked with the signature.
3. What changed — cycle number 47 on that column (which is no longer the column of batch 1), the buffer from the new supplier, the diafiltration run one degree lower. Every change in a spreadsheet, a label, a maintenance report.

Three islands, none of them talking to the others. The drift lives in the joint between the three — and nobody was there listening in time.

How it fits the IRIS system

iLEAN does not replace the ÄKTA — it puts a brain on top that compares curves.

Downstream control is not a problem of missing sensors. The modern biopharma plant already logs in-line UV, conductivity, pH and pressure by the second. The problem is that nobody compares today's real curve with the signature of the good batch and nobody cross-checks the drift with what changed. iLEAN is the putty that seals that crack, without asking you to change the ÄKTA, the TFF skid or the historian.

Edge watches the column and the skid with its own sensors when the SCADA is late. Connect captures the whole chromatograph curve and the buffer supplier's emails. The agent compares against the good batch and anticipates. The person signs — never the other way round.

The three iLEAN pieces applied to protein purification:

• Edge — a terminal with its own pressure, conductivity and temperature sensors over the TFF skid or the chromatography line when the equipment's SCADA is slow or does not export in time. It detects the onset of fouling, pressure drift, cavitation. It works with no network.
• Connect — hooks into the existing ÄKTA, BioSC or chromatograph and captures the full curve second by second. And it also captures the external stuff: the certificate of analysis of the buffer, the email from the resin supplier about the new batch, the column usage report (cycle number 47, no longer the column of batch 1). All of it at second zero.
• Agent — trains on the curves of the batches that came out right and compares today's real curve against that signature. If it detects that the elution peak is going to come out 8% lower or that the TFF pressure will cross the threshold in 20 minutes, it proposes diverting the flow, extending the wash or changing the fraction collected. The person validates; the skid does not adjust itself.

See the full IRIS architecture →

Before and after

Classic downstream vs. downstream anticipated with iLEAN

Aspect	SCADA + assay afterwards	With iLEAN Edge + Connect + Agent
Drift detection	In the final purity assay	Against the good batch signature, live
TFF fouling	When pressure crosses the threshold	Trend detected 20+ min earlier
Column cycle of use	In the column owner's spreadsheet	Connect cross-checks it with the expected curve
Buffer / resin change	Assumed equivalent	Capture of the CoA and the supplier's email
File for inspection	Rebuild from 4 systems	Dossier per batch with a signed curve
Operation with no network	n/a	Edge keeps capturing on the panel's own power

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Biopharma plant producing mAbs or recombinant proteins, downstream with capture chromatography + ion exchange + TFF, multi-product in campaigns.
• Connect + Agent pilot capturing from the existing chromatography system + the MES recipe. First value expected within a few weeks: anticipation of yield or fouling incidents ≥ 30%.
• Indicative payback between 4 and 9 months, driven by the cost of one deviated protein batch and by extended column lifetime through anticipated maintenance.
• The hard lever: a single batch anticipated. With therapeutic proteins, the cost of one batch justifies the system on a single occurrence.

And the process director's reasonable doubt

“What if the AI proposes diverting the flow at the wrong moment and I compromise the peak?” — hallucination is a problem of free generation, not of anchored tasks. Here the AI is anchored to the good batch curve and to the recorded CPPs: in tasks like that, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: the agent proposes, the person signs with their credential — the three rings guarantee that the live process (ring 1, the OT network) only accepts what has been validated and signed. And the aggregate effect in more digitalized plants is striking: the sectors that invested in capturing reality in full raised productivity by as much as 40% compared with the least digitalized ones ^[2].

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks. ^[2] Fundación BBVA / Ivie — productivity through digitalization, 2000-2021.