[ STARTUPS · DRUG DISCOVERY & BIOTECH AI ]
When does GxP start to apply to discovery work?
Not on day one — but the records you keep on day one decide how much of your discovery work can later stand up in an application. Here is where the regulations draw the line, quoted, and what FDA and EMA say about AI used in discovery.
[ WHERE THE REGULATIONS DRAW THE LINE ]
From exploratory research to the first clinical investigation.
Each quotation is verbatim from the regulation or guideline named beneath it. The line headed “What this means” is SPEQ’s interpretation, not regulatory text.
Exploratory research is outside GLP
“The term does not include basic exploratory studies carried out to determine whether a test article has any potential utility or to determine physical or chemical characteristics of a test article.”
Screening, hit-finding and early characterisation are not GLP studies. That is a statement about which regulation applies, not permission to keep poor records: the data may still end up in front of a regulator.
GLP starts with safety studies that support an application
“This part prescribes good laboratory practices for conducting nonclinical laboratory studies that support or are intended to support applications for research or marketing permits for products regulated by the Food and Drug Administration, including food and color additives, animal food additives, human and animal drugs, medical devices for human use, biological products, and electronic products.”
The trigger is purpose, not place: a safety study intended to support an IND is a GLP study whether it runs in-house or at a CRO. Decide which studies those are before they start, because GLP cannot be added afterwards.
Primary pharmacology usually is not GLP
“Such studies are generally conducted during the discovery phase of pharmaceutical development and as such, are not generally conducted in accordance with Good Laboratory Practices (GLP).”
Efficacy and mechanism work from discovery can support a first-in-human programme without being GLP. The same guideline expects the general toxicity studies to be.
The IND has to say which studies were GLP
“For each nonclinical laboratory study subject to the good laboratory practice regulations under part 58, a statement that the study was conducted in compliance with the good laboratory practice regulations in part 58, or, if the study was not conducted in compliance with those regulations, a brief statement of the reason for the noncompliance.”
A study that should have been GLP and was not has to be explained in the application. Knowing which of your studies fall under Part 58 is a question to settle when you plan the IND-enabling package, not when you write it.
Clinical work waits for an IND in effect
“A sponsor shall not begin a clinical investigation subject to § 312.2(a) until the investigation is subject to an IND which is in effect in accordance with § 312.40.”
GCP arrives with the first clinical investigation. ICH E6(R3), the current GCP guideline, was adopted on 6 January 2025 and took effect in the EU on 23 July 2025.
[ WHICH STUDIES MUST BE GLP ]
A short decision list, in the sources’ own terms.
Part 58 turns on purpose: a safety study meant to support an application. ICH M3(R2) and S7A say which study types that usually covers. Where a source says “generally” or “ordinarily”, so does the list.
GLP
An in vivo or in vitro study of the test article’s safety, intended to support an IND, NDA or BLA
Part 58 covers nonclinical laboratory studies that "support or are intended to support applications for research or marketing permits", and the IND, the NDA and the BLA are each named as such an application. (21 CFR 58.1(a); 58.3(d); 58.3(e)(5), (6) and (13) ↗)
GLP
Repeated-dose (general) toxicity studies supporting clinical trials
ICH M3(R2) treats these as the GLP studies: acute toxicity data "can be obtained from non-GLP studies if clinical administration is supported by appropriate GLP repeated-dose toxicity studies". (ICH M3(R2) §4 ↗)
GLP
Safety pharmacology core battery (cardiovascular, central nervous and respiratory systems)
The core battery "should ordinarily be conducted in compliance with GLP"; follow-up and supplemental studies to the greatest extent feasible. (ICH S7A §2.11; ICH M3(R2) §2 ↗)
IT DEPENDS
Acute or single-dose toxicity information
It can come from non-GLP studies when GLP repeated-dose studies support the clinical dosing. Where single-dose studies are the primary support for a human study (for example, microdose trials) they "should be performed in compliance with GLP". (ICH M3(R2) §4 ↗)
IT DEPENDS
A preliminary embryo-fetal study used to support including women of childbearing potential
It "should be conducted under high-quality scientific standards with data collection records readily available or under GLP conditions". (ICH M3(R2) Note 4 ↗)
NOT GLP
Primary pharmacodynamic studies: mode of action and effect on the therapeutic target
Generally conducted during discovery and "not generally conducted in accordance with Good Laboratory Practices (GLP)". S7A: they "do not need to be conducted in compliance with GLP". (ICH M3(R2) §2; ICH S7A §2.11 ↗)
NOT GLP
Basic exploratory studies of potential utility, or of physical or chemical characteristics
Excluded from the definition of a nonclinical laboratory study. (21 CFR 58.3(d) ↗)
NOT GLP
Studies in human subjects, clinical studies, and field trials in animals
Also excluded from the definition. A clinical investigation instead needs an IND in effect before it begins. (21 CFR 58.3(d); 312.20(b) ↗)
Whatever the verdict, the application must account for it: “The NDA or ANDA does not contain a statement for each nonclinical laboratory study that the study was conducted in compliance with the requirements set forth in part 58 of this chapter, or, for each study not conducted in compliance with part 58 of this chapter, a brief statement of the reason for the noncompliance.”
[ AI IN DISCOVERY ]
What the two regulators say — and do not say.
FDA’s AI draft guidance leaves drug discovery out of scope altogether. EMA’s reflection paper calls discovery potentially low regulatory impact, until its results become part of the evidence. Neither says GLP applies to discovery AI as such.
“While the drug product life cycle generally also includes drug discovery, the use of AI for the purposes of drug discovery is not in the scope of this guidance and therefore is not included in our use of the term drug product life cycle.”
“The application of AI/ML in the process of drug discovery can be low regulatory impact if non-optimal performance only affects the developer. However, if results contribute to the total body of evidence presented for regulatory review, principles for non-clinical development (see below) should be followed.”
[ KEEP FROM THE FIRST EXPERIMENT — SPEQ GUIDANCE ]
Four things that cost little now and cannot be rebuilt later.
Research records that can be reconstructed
Who ran what, when, on which material, with which result — attributable and dated at the time. A finding that later supports an application is only as good as the record behind it.
Material and data provenance
Where each compound, cell line, sample and dataset came from, and every transformation since. Provenance cannot be rebuilt after the people and systems have moved on.
Model lineage, if AI produced the result
The training data, model version and settings behind any AI output you rely on. If that output reaches a regulator, EMA expects the principles for non-clinical development to be followed.
A line drawn early between exploratory and GLP work
Name the studies that will support the IND before they start, and place them with a GLP-compliant facility. The rest can stay exploratory — on purpose, and on the record.
Orientation only. SPEQ does not determine whether a study is subject to GLP, whether an IND is needed, or how a regulator will treat AI-generated evidence; that answer comes from the agency and your qualified advisers. Regulatory text checked against the primary sources on 4 October 2026; the GLP decision list on 12 October 2026.
NEXT IN THE DRUG-DISCOVERY JOURNEY
- NEXT STAGEAfter the IND: CMC, Phase 1 to NDA or BLA →What FDA expects at each stage once the IND is in effect, the documents it produces, and where it goes wrong.
- STARTUP GUIDEThe evidence map before your next experiment →
A free account keeps what you save, remembers where you are across devices, and can send you the weekly regulatory briefing. Create a free account