🔷 Regulation of AI Drug Discovery Validation Requirements in China

1. Introduction

Artificial intelligence (AI) is increasingly used in China for drug target identification, molecular screening, drug design, toxicity prediction, and clinical-trial optimization. However, an AI-generated prediction does not automatically establish that a drug is safe, effective, or suitable for human use.

China regulates AI-assisted drug discovery primarily through its existing pharmaceutical regulatory framework, especially the Drug Administration Law, National Medical Products Administration (NMPA) requirements, Good Laboratory Practice (GLP), Good Clinical Practice (GCP), and technical guidelines for pharmaceutical research and development.

The fundamental principle is that AI may assist drug discovery, but the resulting drug candidate must satisfy the applicable scientific, safety, quality, and clinical validation requirements.

2. Legal and Regulatory Framework

A. Drug Administration Law of China (2019)

The Drug Administration Law provides the principal statutory framework for pharmaceutical research, clinical trials, manufacturing, and marketing authorization.

Article 19: Establishes requirements for clinical-trial applications involving drugs, subject to the applicable statutory procedures.

Article 24: Generally requires approval for marketing drugs in China, subject to statutory exceptions.

Article 25: Governs the review and approval of drug marketing applications.

Article 30: Establishes the marketing authorization holder system and associated responsibilities.

These provisions apply to drugs discovered using AI in the same way that they apply to drugs discovered through conventional methods. AI-based predictions cannot substitute for required clinical evidence or regulatory authorization.

B. NMPA and Center for Drug Evaluation (CDE)

The NMPA supervises drug registration, while its Center for Drug Evaluation evaluates applications and supporting evidence.

The Provisions for Drug Registration (2020) establish the framework for drug-development stages, clinical-trial applications, registration review, and marketing authorization.

AI-generated candidates must therefore progress through the relevant development pathway, including nonclinical studies, clinical trials where required, and a marketing application supported by appropriate evidence.

C. Good Laboratory Practice (GLP) and Good Clinical Practice (GCP)

GLP requirements govern applicable nonclinical safety studies, while GCP requirements govern clinical trials involving human participants.

AI may be used to select compounds, model toxicity, or optimize trial design. Nevertheless, applicable studies must satisfy the relevant standards for scientific validity, documentation, quality assurance, and participant protection.

3. Validation Requirements for AI-Based Drug Discovery

A. Computational Model Validation

Drug developers should establish that an AI model is scientifically reliable for its intended purpose. Relevant checks include:

Accuracy and predictive performance using appropriate datasets.

Independent validation using data not used for model training.

Assessment of bias, data quality, and overfitting.

Reproducibility of predictions and documentation of model versions.

Evaluation of the model's limitations and applicability to the intended drug candidate.

These are important scientific and regulatory considerations, although not every item constitutes a separately codified statutory requirement applicable to every AI drug-discovery system.

B. Laboratory and Preclinical Validation

AI predictions must be supported by appropriate experimental evidence. Depending on the drug and development stage, this may include:

Laboratory confirmation of the proposed biological target.

Experimental confirmation of molecular activity and potency.

Pharmacokinetic and pharmacodynamic assessment.

Toxicity and safety pharmacology studies.

Appropriate animal or alternative-model studies.

Applicable nonclinical studies must comply with relevant regulatory standards. A favorable computational toxicity prediction cannot, by itself, establish that a compound is safe for humans.

C. Clinical Validation

Where clinical trials are required, AI-discovered drugs must satisfy the applicable clinical-trial authorization and ethical requirements.

Clinical development generally examines:

Phase I: Initial safety, tolerability, and pharmacokinetics.

Phase II: Preliminary efficacy and dose selection.

Phase III: Confirmatory safety and efficacy in an appropriate patient population.

The precise development pathway depends on the product, its intended use, existing evidence, and applicable regulatory provisions. AI cannot eliminate legally required clinical evidence simply because its predictions are highly accurate.

D. Data Integrity and Traceability

Developers should maintain appropriate records of training data, model development, computational methods, experimental results, and changes to the model. Where AI-generated evidence supports a regulatory application, the sponsor should be able to explain how the evidence was produced and demonstrate its reliability.

Chinese data protection, cybersecurity, and human genetic resources rules may also apply when AI systems process personal medical records, genomic data, or other protected information.

4. Relevant Cases and Regulatory Precedents

A. He Jiankui Gene-Editing Case (2019)

A Chinese court sentenced He Jiankui to three years' imprisonment for illegal medical practice involving gene-edited babies. The case concerned human gene editing, not AI drug discovery.

Relevance: It demonstrates that scientific innovation does not excuse noncompliance with medical, ethical, and authorization requirements. AI-assisted drug developers must likewise comply with applicable regulatory controls.

B. Guo Bing v. Hangzhou Wildlife World (2021)

This case concerned facial-recognition data and privacy protection. It was not a pharmaceutical dispute.

Relevance: It illustrates the legal importance of protecting sensitive data. This principle may be relevant where AI drug discovery uses identifiable patient records or other sensitive health information.

C. Chinese Drug Registration Review Practice

NMPA and CDE review decisions concerning drug registration provide the more directly relevant regulatory context. Drug applications are assessed against applicable standards for pharmaceutical quality, nonclinical safety, and clinical evidence. However, it would be misleading to characterize ordinary drug-registration decisions as judicial precedents specifically establishing AI-validation rules.

Important qualification: No directly controlling Chinese court judgment specifically establishing a comprehensive validation standard for AI-discovered drugs is identified here. The cases above are illustrative analogies, not binding AI-drug-discovery precedents.

5. Institutional Responsibilities and Enforcement

The NMPA and its provincial counterparts oversee relevant pharmaceutical regulatory activities. Developers, clinical-trial sponsors, research institutions, and marketing authorization holders must meet their respective obligations.

Failure to comply may result in rejection or delay of a drug application, suspension of clinical trials, corrective orders, administrative penalties, or other liability under applicable law. Serious misconduct may also trigger civil or criminal consequences.

6. Conclusion

China does not exempt AI-discovered drugs from ordinary pharmaceutical regulation. The central requirement is that AI-generated hypotheses must be supported by reliable computational evidence, appropriate laboratory and nonclinical studies, and clinical evidence where required.

The key distinction is between validating an AI model and validating the drug itself. Model performance can support a discovery decision, but only appropriate experimental, safety, quality, and clinical evidence can establish whether a drug candidate meets the requirements for regulatory authorization.