Biopharmaceutical production scheduling

Scheduling a biologics plant looks like a factory-scheduling problem and behaves like something stricter. In discrete manufacturing, capacity buys throughput and work-in-progress can wait. Neither holds here, and the reasons are biological and regulatory rather than logistical.

The batch has a duration nobody owns

A mammalian fed-batch culture takes roughly two to three weeks to run, and that duration is set by cell growth, nutrient feeding and product accumulation — not by the plant. Adding bioreactors adds parallel batches; it does not shorten one. Perfusion trades this for a longer continuous run at lower volume. Downstream is faster but not free: chromatography cycles, viral filtration and ultrafiltration each occupy a fixed block, and the buffers they consume must be prepared and held in advance.

Because the culture duration is fixed, the only levers a scheduler has are sequence and overlap. That makes the interesting constraints the ones between batches.

Changeover is a validated event, not a cleanup

In a multiproduct facility, switching from one product to the next requires cleaning to a demonstrated limit on carryover. Acceptance is no longer the old rule-of-thumb criteria but health-based exposure limits — a permitted daily exposure derived toxicologically for each compound, per the EMA’s 2014 guideline on shared facilities. Establishing that limit is a validation exercise; meeting it in operation is a fixed cleaning and sampling procedure with its own hold and analysis time.

The scheduling consequence is precise: changeover duration depends on which product preceded which. That turns the plant into a job-shop with sequence-dependent setup times, a combinatorial problem whose solution space grows explosively with the number of products and suites — which is why campaign scheduling groups batches of one product together, accepting inventory cost to avoid changeovers. Single-use equipment removes cleaning validation from the changeover but replaces it with consumable lead times and a different failure mode.

Intermediates expire

The constraint with no counterpart in ordinary manufacturing is the validated hold time. Harvested culture fluid, intermediate pools and purified drug substance may each be held only for a period supported by stability data at a defined temperature. A schedule that leaves a pool waiting for a free chromatography skid past that window does not merely delay the batch — it destroys it. Time in the queue is therefore a hard feasibility constraint, and any plan must reserve the downstream resource before the upstream step is committed.

Shared utilities are usually the real bottleneck, and they are invisible if the plan counts only bioreactors: buffer preparation and hold vessels, water-for-injection capacity, clean-in-place skids, autoclave slots and the number of qualified operators who can be gowned into a suite at once. Chromatography consumes buffer volumes far larger than the process stream, so buffer vessel occupancy often binds before any process vessel does.

Why the plan is wrong on the day it is made

Titre and step yields vary batch to batch, so the quantity a campaign produces is a distribution, not a number. Contamination or a failed in-process check removes a batch outright. And release is not manufacture: sterility testing takes fourteen days by compendial method, and that queue sits after the plant has finished. Practical scheduling is therefore rolling and stochastic — a plan re-solved against actual yields and actual equipment status, held against a deterministic promise it was never able to keep.