Bioprocess AI glossary: CPP, CQA, PAT, hybrid models and more

CPP: critical process parameter

A CPP is a process parameter whose variability affects a critical quality attribute and therefore should be monitored or controlled. ICH Q8(R2) defines the term. In a bioreactor, whether a parameter is critical depends on the specific product and process evidence; a frequently measured parameter is not automatically a CPP.

CQA: critical quality attribute

A CQA is a physical, chemical, biological or microbiological property that should remain within an appropriate limit, range or distribution to ensure desired product quality. ICH Q8(R2) provides the formal definition. The relevant CQAs and their controls are product-specific.

PAT: process analytical technology

PAT refers to a framework for designing, analyzing and controlling manufacturing through timely measurement of critical quality and performance attributes. An online sensor can be part of PAT, but a sensor alone is not a complete control strategy. FDA's PAT guidance discusses the broader development and quality framework.

Hybrid model

A hybrid model combines mechanistic or engineering relationships with data-driven methods such as machine learning. For example, known mass-transfer behavior can constrain a learned relationship between operating conditions and observed titer. The scientific constraints, fitted data and limits of use should all be explicit.

Bioprocess digital twin

A digital twin is a maintained digital representation of a real process that connects process evidence to models for analysis or prediction. A one-time simulation can be valuable, but a twin needs an ongoing connection to the identity and evidence of the process it represents. The required update frequency depends on its intended use.

Data provenance and lineage

Provenance identifies where a record came from; lineage follows how it was mapped, transformed and used. A report that cites a prediction should let a reviewer find the selected run, source value, transformation and model version. This is more precise than storing a final chart alone.

Soft sensor

A soft sensor estimates a difficult-to-measure variable from other observations and a model. It should be evaluated against measured references, with latency, calibration, uncertainty and failure conditions appropriate to the decision. A soft-sensor estimate should remain distinguishable from an assay result.

Design space and operating range

ICH Q8(R2) gives design space a specific regulatory meaning: a demonstrated multidimensional combination of inputs and process parameters that provides assurance of quality and is proposed by the applicant for assessment. A model's training range or suggested operating window is not automatically an approved design space.

Model uncertainty and applicability

Uncertainty describes how much confidence to place in an estimate under stated assumptions. Applicability asks whether the input resembles cases for which the model has evidence. A prediction outside the evaluated scale, cell line, vessel or process version should be flagged for review rather than displayed with the same confidence as an in-range result.

Human-governed AI

In a governed workflow, AI may propose mappings, analyses or recommendations while authorized people inspect evidence and approve consequential actions. Governance includes access, versioning, traceable changes and clearly bounded actions; it is not established by adding a chatbot to a dashboard.

Common questions

Sources and scope

These references inform the terminology and evaluation approach. They do not certify any software or replace a process-specific validation plan.

• ICH Q8(R2) glossary and pharmaceutical development
• FDA PAT framework
• FDA data integrity guidance