Analytics & PAT

Twenty-one subjects — instruments, automation and the argument that connects them. The first recurring idea: every instrument measures something adjacent to the quantity you actually want. Optical density in a well is not a cuvette measurement; trypan blue reports membrane integrity, not life; a particle count says nothing about whether anything is alive; zeta potential is a model output; a sandwich assay stops responding above a certain concentration. Reading an instrument means knowing which neighbouring quantity you are looking at.

The second: the model, not the probe, is the part that fails. PAT stands on calibration models; the step before the instrument dominates the error budget; and a fermentation with forty sensors has only a handful of real degrees of freedom — redundancy is what turns the process historian into a fault detector.

The third: a quality statement is an argument, not a reading. A passed sterility test is weak evidence about a batch, its fourteen days set by how long an injured cell takes to recover; an endotoxin unit is not a mass.

Start with bioprocess analytics and PAT; the individual instruments are best read as case studies of its calibration problem.

Aseptic isolators The physics of barrier separation and the chemistry of vaporised hydrogen peroxide decontamination — including why the aeration phase, not the kill phase, decides how long an isolator is out of service.
Automated nucleic acid extraction The surface chemistry of chaotropic silica binding and magnetic bead capture, and the four places it fails: low-copy samples, short fragments, tough cell walls and carried-over inhibitors.
Bioprocess data historians and analytics Time-series compression, batch alignment and latent-variable statistics — why multivariate models see deviations that no single alarm limit can, and why they still cannot tell you the cause.
Capillary electrophoresis in biopharmaceuticals Electrophoretic mobility, sieving matrices and isoelectric focusing — the physics behind CE-SDS and icIEF, and the artefacts they introduce into purity and charge-variant results.
Collaborative robots in bioproduction The biomechanics behind power-and-force limiting, how contact energy depends on the arm's pose rather than its payload, and the cleanroom constraints that decide where a cobot can actually stand.
HPLC and UHPLC instruments Band broadening and the van Deemter compromise, the pressure penalty that produced UHPLC, and the detection blind spots that decide what a chromatogram is actually evidence of.
Particle size and zeta potential analysers Brownian motion and the autocorrelation function behind dynamic light scattering, the resolution limits it cannot escape, and why electrophoretic mobility stops yielding a meaningful zeta potential at physiological salt.
Sterility testing systems The sampling statistics that limit what a compendial sterility test can prove, the microbiology behind the fourteen-day incubation, and what rapid alternative methods actually detect instead.
Automated ELISA analyzers The binding and enzyme chemistry underneath automated microplate immunoassay, and the two failure modes — the high-dose hook effect and timing drift — that decide whether an automated result can be trusted.
Automated sample preparation for analytics The physical chemistry of extraction, clean-up and small-volume liquid transfer, and why automating a preparation improves its precision without touching its bias.
Bioreactor gas mixing systems The transport physics of oxygenation in stirred tanks — kLa, bubble coalescence, interfacial shear and dissolved CO2 accumulation — and why the same knobs cannot be held constant at two scales.
Cell counters and viability systems The dye chemistry, impedance physics and fluorescence logic behind automated cell counting — and the Poisson limit that decides how many cells must be seen before a viability figure means anything.
Endotoxin testing systems The enzymatic cascade behind the LAL test, the Factor G cross-reaction that recombinant Factor C removes, and the low endotoxin recovery problem that no reagent chemistry has solved.
Plate readers in bioproduction Beer–Lambert, scattering, inner filter effects and time-resolved gating in microplate photometry — and the statistic that decides whether a plate assay is fit to screen with.
Automated feed systems for bioreactors The metabolic reason fed-batch feeding exists, the open-loop problem created by the absence of a reliable in-line glucose sensor, and the physical limits — osmolality, tubing drift and mixing time — that bound what a feed strategy can do.
Bioprocess analytics and PAT The logic of the FDA's process analytical technology framework, the spectroscopy that makes in-line measurement possible in a living broth, and why a chemometric model trained at bench scale is the weakest joint in the chain.
Biosafety cabinets and laminar flow hoods Why the most penetrating particle size is around 0.3 micrometres, how unidirectional flow protects a product, and why a Class II cabinet's protection of the operator rests entirely on an inflow velocity that gives no visible sign when it fails.
Cleanroom environmental monitoring How optical particle counting works and why its result is an equivalent optical diameter, why counting statistics rather than instrument quality set the limit at the cleanest grades, and why viable monitoring is a separate and much weaker measurement.
Flow cytometers in bioproduction How hydrodynamic focusing and fluorescence detection make single-cell measurement possible, why compensation removes spillover but not the noise it brings with it, and why sorting purity and yield trade against each other by Poisson statistics.
Mass spectrometry in biopharmaceutical characterisation Why multiple charging brings large proteins within reach of ordinary analysers, what resolving power has to deliver to see a single deamidation, and why ion suppression makes a matrix-matched standard non-negotiable for quantitation.
Robotic biobanking and sample storage The cryobiology behind long-term biospecimen storage: why the glass transition and not the freezing point sets the archive temperature, why the two-factor theory of freezing injury makes cooling rate cell-specific, and why automation is a thermal-history argument rather than a throughput one.