Research & Discovery
Research and discovery is where the product exists only as a hypothesis: a biological target worth modulating, a device concept worth prototyping, a formulation worth stabilising, a food product worth developing. For a medicine this means target identification, screening, hit-to-lead chemistry or construct engineering, and candidate selection. For a device it means user-needs capture, concept prototypes, and feasibility work. For food, veterinary, and cosmetic products it means recipe and formulation development, ingredient selection, and early safety reasoning. The defining feature of the phase is that most GxP obligations have not yet attached — and the defining risk is behaving as if that means quality has not yet attached either.
What this page does not claim
Phases are an organising device, not a regulatory mandate — real programmes overlap them, and a medical device, a food product and a small-molecule drug traverse them differently. Each page says where the differences bite.
What happens in this phase
Research and discovery is where the product exists only as a hypothesis: a biological target worth modulating, a device concept worth prototyping, a formulation worth stabilising, a food product worth developing. For a medicine this means target identification, screening, hit-to-lead chemistry or construct engineering, and candidate selection. For a device it means user-needs capture, concept prototypes, and feasibility work. For food, veterinary, and cosmetic products it means recipe and formulation development, ingredient selection, and early safety reasoning. The defining feature of the phase is that most GxP obligations have not yet attached — and the defining risk is behaving as if that means quality has not yet attached either.
The industries diverge early. In devices, the regulatory clock starts sooner than most teams expect: design controls under 21 CFR 820 and ISO 13485 attach once development of a device intended for market begins, so the boundary between "research" and "development" is a controlled decision, not a vibe. In pharmaceuticals, discovery data rarely reaches a regulator directly, but candidate-selection data, developability assessments, and early stability observations become the foundation the entire CMC story is built on. A discovery dataset that cannot be reconstructed — no metadata, no instrument context, no traceable analysis — quietly weakens every downstream claim that rests on it.
The quality function's job here is proportionate, not absent. It sets the data-integrity floor for research records, defines when GxP disciplines will switch on and who decides, and applies risk-based thinking to the choices that will be expensive to revisit — material selection with biocompatibility in mind, an excipient or ingredient with a known supply fragility, a molecule with a foreseeable impurity liability. Increasingly it also governs the machine-learning tools used in discovery: a model that ranks candidates is not itself a GxP system, but the organisation should already be recording what it relied on, because the question "why this candidate" will be asked years later.
- A defined unmet need, target hypothesis, or product concept with a sponsor or portfolio decision behind it.
- Research data-management and record-keeping conventions in place, so early data is attributable and reconstructable even though it is not yet GxP-controlled.
- A stated organisational rule for when formal development — and with it design controls or GLP obligations — begins, and who makes that call.
- For devices: documented user needs and intended use sufficient to reason about classification and the regulatory pathway ahead.
- A nominated candidate or frozen concept, selected against pre-agreed criteria rather than enthusiasm.
- A developability and risk assessment covering manufacturability, stability liabilities, material and supply risks, and — for devices — a first hazard-based look at the design.
- The data package supporting candidate selection is complete, traceable, and retained, because submissions and design history files will later cite it.
- A documented decision to enter formal development, triggering the GxP disciplines that govern the next phases.
SPEQ synthesis. Phase boundaries and gate criteria are an organising device for planning and review, not a regulatory mandate. Real programmes overlap phases and re-enter them; treat these as the questions worth answering, not a compliance checklist.
- Define the research data-integrity floor — attributability, original records, traceable analysis — before any GxP mandate forces it.
- Own the boundary decision: the documented trigger at which design controls, GLP, or formal development quality systems switch on.
- Bring quality-risk-management thinking into candidate and material selection, where a cheap early decision prevents an expensive late one.
- Set expectations for records retention of discovery data that will underpin later regulatory claims.
- Advise on the governance of AI and machine-learning tools used in discovery, so reliance on them is recorded and explainable.
- Candidate-selection report with the supporting screening and characterisation data
- Target product profile or, for devices, documented user needs and intended use
- Developability and early risk assessment (manufacturability, stability, materials, supply)
- Laboratory notebooks and research records, retained and reconstructable
- Intellectual-property and freedom-to-operate position
- The documented decision record marking entry into formal development
- Treating "not yet GxP" as "no rules at all", producing discovery data that cannot later support the claims built on it.
- Devices drifting from research into development without invoking design controls, so the design history file starts with a hole a regulator will find.
- Candidate selection driven by potency alone, deferring manufacturability and stability liabilities that surface at scale-up as redesign.
- No record of what machine-learning or literature tools were relied on for decisions later cited in submissions.
- Losing early data to informal storage — personal drives, unmanaged instruments — so the provenance chain breaks at its first link.
Derived from the 5 standards SPEQ maps to this phase, across 4 regulatory bodies: FDA, ICH, ISO, MHRA.
FREQUENTLY ASKED
Does GxP apply during research and discovery?
Mostly not yet — and precisely where it starts is the decision that matters. Discovery science is generally outside GLP, GMP, and GCP. But the boundary is real and consequential: GLP attaches when a study is intended to support safety assessment for a regulatory submission, and device design controls attach when development of a device intended for market begins. Well-run organisations define those triggers in advance and document the crossing, because a regulator reviewing a submission or design history file will look for the point at which controlled development began — and for evidence that nothing load-bearing predates it unexamined.
Why should quality care about data no regulator will ever inspect?
Because some of it will be inspected — just much later, wearing a different label. Candidate-selection data reappears in the development history a submission narrates; early stability observations shape shelf-life reasoning; discovery-era material choices become the starting point of the control strategy. If that data is unattributable or unreconstructable, the downstream document inherits the weakness. The practical standard is proportionate: not validation, but attributable records, original data preserved, and analysis that can be traced from raw signal to conclusion — the ALCOA+ floor applied with a light hand.
When do design controls start for a medical device?
When development of a device intended for commercial distribution begins — not when the team feels ready. Both 21 CFR 820 and ISO 13485 tie design controls to the development of the device, and regulators expect the design history file to begin with design planning and user needs, not to be reconstructed backwards from a prototype that already exists. Research and feasibility work sits outside design controls, which is exactly why the transition should be a documented decision: it draws the defensible line between exploration, which is free, and development, which is controlled.