Chapter 1
A map of how living cells are turned into medicines and other biological products, the handful of numbers that describe a production run, the regulatory ideas that shape every decision, and an honest inventory of what the WIA-BIO-015 standard actually contains.
Before a reader can use a data standard for bio-manufacturing, it helps to know what is being described. A standard that encodes "cell density", "titer" or "Protein A step recovery" only makes sense to someone who knows why those quantities matter, where in the process they are measured and which decisions they drive. This first chapter therefore does two jobs. It lays out the physical process — from a frozen vial of cells to a filled container of purified product — and it maps the parts of WIA-BIO-015 onto that process, so that every later chapter can be read as a zoom into one region of the same picture.
Throughout the book, WIA-BIO-015 means the specification spec/WIA-BIO-015-v1.0.md (version 1.0.0, published 2025-12-26) together with its reference TypeScript SDK and command-line tool. Where the specification and the SDK disagree, the text says so explicitly rather than smoothing over the difference.
A biologic is a product whose active substance is made by a living system rather than assembled by synthetic chemistry. A small-molecule drug such as aspirin can be described completely by its chemical structure and made by any competent route that reaches that structure. A monoclonal antibody, by contrast, is a glycoprotein of roughly 150 kDa assembled from two heavy and two light chains, folded by the cell's own machinery and decorated with sugar chains whose composition varies from molecule to molecule. What leaves the bioreactor is not one molecule but a population of closely related variants: different glycoforms, charge variants, small amounts of aggregates and fragments.
The distribution of those variants depends on the cell line, the medium, the temperature and pH history, the feeding strategy, the harvest timing and the purification sequence. Change any of them and the product may change in ways that affect potency, half-life or immunogenicity. This is the origin of the industry maxim "the process is the product": for a biologic, the manufacturing process is part of the product's identity, and regulators expect a manufacturer to show that any change to the process leaves the product comparable (the ICH Q5E guideline addresses exactly this comparability question). A data standard for this field therefore cannot stop at recording the final result; it must capture the process conditions that produced it.
WIA-BIO-015 names the classes of product it is meant to cover in the ProductType union of the SDK. The list is broader than therapeutic proteins, and it is worth reading each entry as a statement about the process rather than the molecule:
ProductType value | What is being manufactured | What makes the process distinctive |
|---|---|---|
antibody | Monoclonal antibodies and related formats (bispecifics, Fc-fusions) | Mammalian culture, Protein A capture, tight glycosylation control |
protein | Other recombinant therapeutic proteins (hormones, cytokines, blood factors) | Host choice depends on whether glycosylation and complex folding are needed |
enzyme | Therapeutic or industrial enzymes | Activity assays rather than binding assays define potency; often microbial or yeast hosts |
vaccine | Subunit antigens, virus-like particles, inactivated or live viral vaccines | Potency and titer units are product-specific, not always g/L |
viral-vector | Vectors such as AAV, lentivirus or adenovirus used to deliver genes | Often transient transfection or infection of HEK-293-type cells; titers expressed as particles or genome copies |
cell-therapy | Living cells as the final product | No purification of a molecule; the cells themselves must stay viable and functional |
gene-therapy | Genetic payloads and their delivery vehicles | Overlaps with viral-vector manufacture; empty/full capsid ratio is a key quality question |
biofuel | Fuels and chemicals from microbial fermentation | Non-GMP, cost-driven; yield on substrate dominates the economics |
custom | Anything not listed | Escape hatch for implementations; carries no semantics |
The weight of the standard, however, is clearly on protein products made in suspension culture and purified by chromatography — the monoclonal antibody is the reference case in the specification's worked examples, its downstream section and its monitoring thresholds. Cell and gene therapies appear in the type system but not yet in the calculation rules; Chapter 8 returns to what would be required to cover them properly.
The first engineering decision in any biologic programme is the choice of host organism, called the expression system. The host determines which post-translational modifications are possible, how fast cells divide, which impurities must later be removed and which viruses the process must be shown to clear. WIA-BIO-015 enumerates the supported hosts in the ExpressionSystem union: CHO, E.coli, Yeast, Pichia, Insect, HEK-293, PER.C6 and custom.
| Expression system | Glycosylation | Growth (SDK reference value) | Main strengths | Typical products |
|---|---|---|---|---|
| CHO (Chinese hamster ovary) | Complex, largely human-compatible N-glycans | μ ≈ 0.03 h⁻¹, so td ≈ 23 h | Secretes correctly folded, glycosylated protein; long regulatory track record; adapted to serum-free suspension culture | Monoclonal antibodies, Fc-fusion proteins, complex glycoproteins |
| E. coli | None | μ ≈ 0.7 h⁻¹, so td ≈ 1 h | Very fast growth, cheap media, simple genetics | Non-glycosylated proteins, antibody fragments, peptides; often expressed as insoluble inclusion bodies that must be refolded |
| Yeast / Pichia pastoris | High-mannose type, not human-like unless the strain is engineered | μ ≈ 0.25 h⁻¹, so td ≈ 2.8 h | Eukaryotic folding and secretion with microbial-like speed; methanol-inducible promoters in Pichia | Enzymes, hormones, some vaccine antigens |
| Insect cells (baculovirus system) | Simpler, mostly paucimannose glycans | No SDK constant | Good for complex multi-subunit assemblies | Virus-like particles, some vaccines, viral-vector production |
| HEK-293 | Human glycosylation | No SDK constant | Human cell line; readily transfected; supports replication of many viral vectors | AAV and lentiviral vectors, some therapeutic proteins |
| PER.C6 | Human glycosylation | No SDK constant | Human retina-derived line developed for adenovirus production | Adenoviral vectors and vaccines |
The growth figures in this table come from BIO_CONSTANTS.GROWTH_RATE in the SDK (CHO 0.03, E. coli 0.7, yeast 0.25 h⁻¹), and the doubling times follow from td = ln 2 / μ. They are reference values, not guarantees. A fast E. coli strain in rich medium at 37 °C can double in well under an hour, and a CHO clone's growth rate varies with the clone, the medium and the process phase. Chapter 2 derives these relationships properly.
The specification's own table (§2.1.1) adds a "typical yield" column: 1–10 g/L for CHO, 5–50 g/L for E. coli, 1–15 g/L for Pichia, 0.1–5 g/L for insect cells, and 10⁶–10⁹ vp/mL (virus particles per millilitre) for HEK-293. Three caveats apply when reading it. First, these are the standard's illustrative ranges, not survey data, and a given product can land far outside them. Second, the columns are not comparable across rows: a microbial titer usually refers to total expressed protein, much of which may be insoluble and lost during refolding, whereas a CHO titer refers to secreted, correctly folded product. Third, the HEK-293 entry is expressed in particles rather than grams, which is a reminder that for vectors the useful quantity is often the number of infectious or full particles, not their mass.
In practice, the choice of host follows from the molecule. If the product needs human-like glycosylation to be active or to have an acceptable half-life — as an antibody's Fc region does, with its conserved N-glycan at asparagine 297 — a mammalian host is required. If the product is a small, non-glycosylated protein, the speed and cost of a microbial host usually win. Viral vectors follow the virus: the producer cell must supply whatever the vector needs to replicate or assemble.
Almost every protein biologic is made by the same sequence of unit operations, even though the details differ from product to product. The sequence is conventionally divided into upstream processing (USP), where cells grow and make product, and downstream processing (DSP), where the product is separated from cells, medium and impurities. Quality control runs alongside both.
Production starts from a cell bank. A master cell bank (MCB) is a large set of identical frozen vials made from the selected clone; a working cell bank (WCB) is derived from it and used for routine production. Both are characterised for identity, stability and freedom from adventitious agents (ICH Q5D covers the derivation and characterisation of cell substrates). Because every batch starts from the same bank, the bank is the first guarantee of consistency.
A thawed vial holds far too few cells to seed a production vessel directly, so the cells are expanded through a seed train: a sequence of progressively larger vessels, from flasks through rocking bags to one or more seed bioreactors. The specification (§2.3.1) describes a typical progression from a 25 cm² T-flask to a production reactor of 1,000–10,000 L with a split ratio of 1:5 to 1:10 per step, and an inoculation density of 2–5 × 10⁵ cells/mL for mammalian cultures (§2.3.2). The seed train is a scheduling problem as much as a biological one: each step must be transferred while cells are still growing exponentially.
The production bioreactor is where most of the product is made. It is usually a stirred tank, either stainless steel or single-use, controlled for temperature, pH, dissolved oxygen, agitation and pressure. It may be run in batch, fed-batch or perfusion mode, which the SDK encodes as OperationMode (batch, fed-batch, perfusion, continuous). For CHO antibody production, fed-batch culture of roughly two weeks is the most common mode; Chapter 3 compares the modes in detail.
At the end of the culture comes the harvest. For secreted products the cells are waste at this point, and the goal is to remove them and their debris without damaging the product. Continuous disc-stack centrifugation followed by depth filtration is a common combination; small single-use processes may use depth filtration alone.
The clarified harvest contains the product at a few grams per litre alongside host cell proteins (HCP), DNA, media components and potentially viruses. Downstream processing removes these in orthogonal steps, each exploiting a different physical property:
The specification covers harvest through UF/DF in §5 and gives the formulas behind each step; Chapter 4 works through them. What matters for the overall picture is that each step has a recovery below 100 %, and step recoveries multiply. Illustratively, three steps at 95 %, 92 % and 90 % together keep only about 79 % of the product — the reason why process development tries to remove steps rather than add them.
The line between USP and DSP is organisational as much as technical. In Figure 1-1 it is drawn after harvest, which matches the way the SDK separates bioreactor types (BioreactorStatus, CellCultureMetrics) from purification types (ChromatographyStep, PurificationProtocol). Some companies assign clarification to the downstream team because it shares equipment and skills with filtration. In a data standard, the important thing is not where the line is drawn but that one object — here a batch identifier — links both halves, so that a quality result can always be traced back to the culture that produced it.
The README and specification (§1.4) summarise a production run with four figures: yield, volumetric productivity, specific growth rate and titer. They are simple to calculate, and for that reason it is easy to misread them. The table below gives each definition as the standard states it, and then what the number can and cannot tell you.
Yield Y = (P_final × V_final) / (S_initial × V_initial) [dimensionless]
Volumetric productivity Q_p = (P_final − P_initial) / t [g/L/h]
Specific growth rate μ = ln(X_2 / X_1) / (t_2 − t_1) [h⁻¹]
Titer = product concentration at harvest [g/L]| KPI | What it measures | What it does tell you | What it does not tell you |
|---|---|---|---|
| Yield Y | Mass of product at harvest divided by mass of substrate supplied | How much of the carbon source ended up as product, a rough economic indicator for substrate-limited processes | It is not a stoichiometric yield: most glucose in animal-cell culture goes to biomass, energy and lactate, and the definition ignores all of that. It says nothing about purity or recovery. |
| Volumetric productivity Qp | Titer gained per unit volume per unit time | How well the reactor volume is used; the basis for comparing processes of different duration | Whether cells are productive individually (that is qp, Chapter 2); turnaround and cleaning time between batches, unless t is taken as the whole cycle |
| Specific growth rate μ | Exponential rate of increase of cell density | Culture health during growth; basis for doubling time and feed design | Anything outside the exponential phase; if total rather than viable cells are used, it hides cell death |
| Titer | Product concentration in the harvest | The single most quoted upstream result; drives downstream column sizing | Product quality. A higher titer with more aggregates or wrong glycoforms can be a worse result. |
The yield definition deserves a careful look, because it differs from the textbook yield coefficient YP/S, which relates product formed to substrate consumed. WIA-BIO-015 relates product mass at harvest to substrate mass supplied at the start. The specification's example (Appendix A.1) uses a product concentration of 5.2 g/L in 1,000 L and an initial substrate concentration of 50 g/L in 1,000 L:
Total product = 5.2 g/L × 1,000 L = 5,200 g
Total substrate = 50 g/L × 1,000 L = 50,000 g
Y = 5,200 / 50,000 = 0.104 (10.4 %)Two details show why the definition must be read literally. First, the example assumes the final volume equals the initial volume, which is only true for a batch culture without feeds. Second, in a fed-batch process a large share of the glucose arrives in the feed, not in the starting medium. The SDK handles this better than the specification's formula: its calculateYield accepts optional feedVolume and feedSubstrate fields and adds feedSubstrate × feedVolume to the denominator. As an illustrative example, if the same run also received 100 L of feed containing 400 g/L glucose, the substrate supplied would rise to 90,000 g and the yield would fall to about 0.058 — the product did not change, only the accounting. When two sites compare yields, they must agree on whether the feed is counted.
The SDK also grades the result: above 0.15 it reports excellent, above 0.10 good, above 0.05 fair, and otherwise poor. These are the standard's grading bands for its own yield definition; they are not industry benchmarks, and Chapter 6 treats them as such.
Volumetric productivity divides the titer gained by the culture time. Appendix A.2 reaches 5.2 g/L in 240 h, which gives 0.0217 g/L/h, or 0.52 g/L/day. The daily form is easier to compare because culture durations are usually counted in days. Note what t means: if it is the time from inoculation to harvest, it ignores the days spent cleaning, sterilising and refilling the vessel. A facility planner interested in annual output would divide by the full cycle time instead.
The specific growth rate uses the natural logarithm because cells in the exponential phase grow in proportion to their own number: X(t) = X0eμt. For the CHO reference value μ = 0.03 h⁻¹, the doubling time is ln 2 / 0.03 ≈ 23 h. Measured over the wrong window — for instance across the transition into the stationary phase — the same formula returns a meaningless average. Chapter 2 shows how to choose the window and how to fit μ by regression rather than from two points.
Titer is simply concentration, but it is the number that couples upstream to downstream. A doubling of titer at the same volume doubles the mass the capture column must bind, which can double its size or its number of cycles. Titer improvements therefore shift the process bottleneck from the bioreactor to purification — a recurring theme when a process is intensified.
Biologics for human use are manufactured under a set of regulatory expectations that influence every data field in WIA-BIO-015, even where the standard does not mention them. The goal here is not a legal summary but an explanation of why the standard talks about critical process parameters, critical quality attributes and batch records at all.
| Framework | The question it answers | Where it appears in this book |
|---|---|---|
| Good Manufacturing Practice (GMP) — e.g. US 21 CFR Parts 210/211 and 600-series, EU GMP (EudraLex Volume 4) with its Annex 2 on biological substances | Is the product made in a controlled, documented, traceable way? | Batch records and data integrity (Chapter 7) |
| ICH Q8(R2) Pharmaceutical Development and the Quality by Design (QbD) approach, with ICH Q9 (quality risk management) and ICH Q10 (pharmaceutical quality system) | Do we understand how process parameters affect product quality, and have we defined a design space? | CQAs (Chapter 4), DOE and capability (Chapter 8) |
| ICH Q11 Development and Manufacture of Drug Substances | How is the drug-substance process developed and its control strategy justified? | Process description types (Chapter 5) |
| FDA guidance "Process Validation: General Principles and Practices" (2011) | Is the process capable of consistently delivering quality product over its life? | Validation lifecycle (Chapter 8) |
| FDA guidance on Process Analytical Technology (PAT, 2004) | Can quality be assured by measuring and controlling the process in real time? | Monitoring and control (Chapters 3 and 7) |
| ICH Q5A (viral safety) and ICH Q5D (cell substrates) | Are the cell bank and the process demonstrably safe with respect to viruses and adventitious agents? | Cell banks (this chapter), viral clearance (Chapter 4) |
The common thread is a chain of reasoning. A critical quality attribute (CQA) is a property of the product — purity, aggregate level, glycan profile, potency — that must stay within limits to ensure safety and efficacy. A critical process parameter (CPP) is a process variable — pH, temperature, dissolved oxygen, feed rate — whose variation affects a CQA and therefore must be monitored or controlled. QbD asks the manufacturer to establish these links experimentally; process validation asks for evidence that the controlled process keeps delivering; PAT asks for measurements fast enough to act on. The WIA-BIO-015 data model can be read as a vocabulary for exactly this chain: CultureConditions and SensorReading hold CPPs, CriticalQualityAttribute and QualityControlResult hold CQAs, and ProcessCapability and ValidationResult hold the evidence.
One widespread simplification needs to be flagged now. Both the README and the specification (§8.1.2) state that process validation requires three consecutive successful batches. The 2011 FDA guidance does not prescribe a fixed number of batches; it asks the manufacturer to justify the number from process knowledge and variability. "Three batches" is a historical convention. Chapter 8 discusses how to replace it with a statistically argued plan.
With the process in view, we can now look at the standard itself. Its main text is a single specification document of ten sections and one appendix:
| Section | Content | Covered in |
|---|---|---|
| §1 Introduction | Purpose, scope, philosophy, six core terms | Chapter 1 |
| §2 Upstream Processing | Expression systems, clone selection, media, feed strategy, seed train, inoculation | Chapters 1–2 |
| §3 Bioreactor Systems | Reactor types, operating modes, CPPs (pH, temperature, DO) | Chapter 3 |
| §4 Cell Culture Optimization | Growth kinetics, nutrient consumption, product formation | Chapter 2 |
| §5 Downstream Processing | Harvest, chromatography, viral inactivation, UF/DF | Chapter 4 |
| §6 Process Analytical Technology | Online, at-line and spectroscopic monitoring; PID and advanced control | Chapter 7 |
| §7 Quality by Design | CQAs, DOE, capability indices | Chapters 4 and 8 |
| §8 GMP Compliance | Batch records, validation, cleaning validation | Chapters 7–8 |
| §9 Implementation Guidelines | Required components, API interfaces, error codes — the normative core | Chapters 5–6 |
| §10 References and Appendix A | Literature, constants, related standards; two worked calculations | Chapters 1, 6, 8 |
Sections 1 to 8 are largely explanatory: they describe established bioprocess science. The part that turns the document into a standard is §9.1, which states that any compliant system must include five components:
calculateYield and calculateProductivity.monitorBioreactor, returning a MonitoringResult with a status of optimal, acceptable, warning or critical.optimizeFermentation and the DOEConfig types.designPurification, returning a PurificationProtocol.calculateProcessCapability and validateBatch.Section 9.2 then gives interface sketches for the first, second and fourth components, and §9.3 defines six error codes, B001 (low cell viability) through B006 (equipment failure), each paired with an operator action. The SDK extends this list to ten codes by adding B007 (invalid parameters), B008 (GMP violation), B009 (out of specification) and B010 (insufficient data). Chapter 6 treats the error model in full.
A reader implementing the standard needs to know which files carry weight. The repository contains more documents than the specification alone, and they are not all of equal value:
| Artefact | What is inside | Maturity and how this book uses it |
|---|---|---|
spec/WIA-BIO-015-v1.0.md | The ten-section specification described above | The only substantive specification. Its §9 is normative. Several formulas in §2–§8 need correction; each is pointed out where it is used. |
api/typescript/src/types.ts | About 1,000 lines of TypeScript interfaces, enums, constants and the error class | The most detailed definition of the data model. Where it differs from §9.2, this book treats the SDK shape as the de-facto wire format (Chapter 5). |
api/typescript/src/index.ts | The BioManufacturingSDK class and standalone functions | A working reference implementation of the calculations. Some functions return fixed placeholder values rather than computed results (Chapter 6). |
api/typescript/package.json | Package @wia/bio-015 built with tsup to CommonJS, ESM and type declarations | A build definition. Its presence does not mean the package is published to a public registry. |
cli/wia-bio-015.sh | Bash tool with calc-yield, calc-productivity, monitor, design-purification, optimize and version | Useful for quick checks; several options differ from the README examples (Chapter 6). |
spec/PHASE-1 … PHASE-4 | Four documents titled data format, API interface, protocol and integration | Generic template text shared across the WIA series and headed with the wrong standard ID. Used only as non-normative background. |
simulator/index.html | A generic five-tab web interface (overview, configuration, simulation, results, documentation) | Not a bioreactor model. This book does not rely on it for any result. |
This map sets the tone for the rest of the book. WIA-BIO-015 v1.0 is a useful skeleton: its vocabulary of types is sensible, its core formulas are correct, and its reference code can be run and tested. It is also, in places, a placeholder. An engineer who treats every number in the repository as authoritative will build a system that reproduces placeholders; one who knows where the placeholders are can build a system that is correct and still conforms. Chapter 8 collects these gaps into a list of proposed errata.
The README names WIA-DATA, WIA-INTENT, WIA-OMNI-API and WIA-SOCIAL as integration partners, and §10.3 lists further related WIA standards in the biology series. None of these interfaces is specified in WIA-BIO-015 v1.0 itself; they are named integration points. Chapter 7 shows how the standard's objects fit into the plant systems that already exist — control systems, historians, manufacturing execution systems (MES) and laboratory information management systems (LIMS) — which matters far more in practice than links to other WIA documents.
The eight chapters follow the path of the product and then the path of an implementation. Chapters 2–4 cover the science of upstream and downstream processing; Chapters 5–6 cover the standard's data model and interfaces; Chapters 7–8 cover operation, integration and validation.
| Chapter | Topic | Main outcome for the reader |
|---|---|---|
| 1 | Landscape and anatomy of WIA-BIO-015 | A map of the process and of the standard |
| 2 | Upstream: cell lines, media, growth kinetics | Compute μ, td, qp and IVC from culture data |
| 3 | Bioreactor operation and CPPs | Reason about mixing, oxygen transfer and operating modes |
| 4 | Downstream processing and quality attributes | Size a capture column, plan diafiltration, list the CQAs of a mAb |
| 5 | The data model | Build valid BioreactorStatus and QualityControlResult objects |
| 6 | SDK and CLI | Reproduce the reference calculations and interpret error codes |
| 7 | Batch protocol and system integration | Place the standard's objects in a real plant architecture |
| 8 | Validation, conformance testing and next steps | Write a test suite and know the gaps of v1.0 |
Prerequisites. Readers from a bioprocess background will find Chapters 5 and 6 require only basic familiarity with JSON and TypeScript interfaces. Readers from a software background will need no biology beyond this chapter, but should be comfortable with exponentials and logarithms, since most process quantities are derived from exponential growth or first-order decay.
Notation and units. The book follows the units used in the SDK field comments, because those are what an implementation will exchange:
| Quantity | Symbol | Unit used in this book | Note |
|---|---|---|---|
| Product concentration, titer | P | g/L | mg/L appears for low-titer products; always state which |
| Cell density | X | cells/mL | Viable cell density unless stated; microbial processes often use g/L dry cell weight or OD instead |
| Specific growth rate | μ | h⁻¹ | Some literature uses day⁻¹; divide by 24 to convert |
| Glucose | S | g/L | Stored in g/L in CellCultureMetrics |
| Lactate, glutamine, ammonia | — | mM | Stored in mM, unlike glucose — a common source of unit errors |
| Dissolved oxygen | DO | % air saturation | Not mg/L; 100 % means equilibrium with air at process conditions |
| Volumetric productivity | Qp | g/L/h or g/L/day | The SDK returns both |
| Specific productivity | qp | pg/cell/day | Defined in Chapter 2 |
Code identifiers such as BioreactorStatus, monitorBioreactor and B001 are written exactly as they appear in the SDK. Numerical examples labelled "illustrative" or "for example" are constructed for teaching and are not measured data; figures attributed to the specification or SDK are quoted as those sources state them, with any known problem noted beside them.
Key points of Chapter 1