How AI in Personalized Medicine and Drug Discovery Works, Is Regulated and Falls Short

From genetic profiles to drug pipelines: what AI is described as doing in medicine, what regulators say and where it falls short

Por Glauber em 22 de October de 2026

What Personalized Medicine Means

The National Human Genome Research Institute (NHGRI) defines personalized medicine as an emerging practice that uses an individual’s genetic profile to guide decisions about prevention, diagnosis and treatment. It can help choose the right drug or therapy, the dose and the schedule, in contrast with one-size-fits-all care. The National Cancer Institute (NCI) uses the term precision medicine for a similar idea. In cancer, biomarker tests are used to pick treatments most likely to help and to avoid those unlikely to work.

Biomarker testing is routine for non-small cell lung, breast and colorectal cancers. Even so, the NCI says precision medicine is not yet routine care for most patients.

Pharmacogenomics: Matching Drugs to Genes

Pharmacogenomics, also called pharmacogenetics, is the part of genomic medicine that uses a patient’s genomic information to guide the choice of drug. According to the NHGRI, pharmacogenetic tests can show whether a drug is likely to be harmful, helpful or ineffective for a given patient.

The FDA publishes a Table of Pharmacogenetic Associations that lists gene and drug pairs where there is sufficient evidence that variants may change drug metabolism, response or the risk of adverse events. One entry pairs abacavir with the HLA-B gene: patients who carry the *57:01 allele have a higher risk of hypersensitivity reactions, and the table says not to use abacavir in them. The FDA adds cautions. Listing a pair does not mean the agency recommends testing before prescribing, unless the test is a companion diagnostic, and most associations have not been evaluated for the effect of testing on clinical outcomes. Patients should not change their medicines without talking to the prescriber.

Where AI Enters the Picture

The clearest example comes from the NCI. In a 2024 Nature Cancer paper, NIH researchers described PERCEPTION, an AI tool that uses single-cell RNA sequencing of tumor cells to predict the response to a specific drug. The models were built from published cell-line drug-screen data for 44 FDA-approved cancer drugs.

The team checked it against published patient data. It covered 41 multiple myeloma patients on a four-drug combination, 33 breast cancer patients on a two-drug combination, and 24 non-small cell lung cancer patients on targeted therapies, where the model predicted the development of resistance. The NCI is clear about the status of this work: it is a proof-of-concept study that used retrospective data, not a prospective trial. PERCEPTION is a proof of concept validated on retrospective data, and single-cell data cost much more than bulk data and are not yet widely available in clinical settings. Related reading: digital therapeutics for personalized healthcare, and the Internet of Bio-Things.

AI in Drug Discovery and Development

The FDA’s Center for Drug Evaluation and Research (CDER) reports more than 500 submissions with AI components between 2016 and 2023 and says that use has grown considerably. Its page offers no newer count. The EMA’s reflection paper cites uses such as reducing animal testing, selecting trial participants and supporting pharmacovigilance.

In structural biology, the Nobel Prize announcement of October 9, 2024 honored Demis Hassabis and John Jumper for predicting protein structures with AI (AlphaFold) and David Baker for computational protein design. The prize does not, by itself, show that a medicine was approved.

How Regulators Are Responding

The FDA published a draft guidance in January 2025 on AI to support regulatory decisions about drugs and biologics. It proposes a risk-based credibility framework with seven steps, anchored in the model’s context of use. The FDA page does not say whether the guidance has been finalized. In January 2026 the EMA and the FDA jointly published ten principles of good practice for AI in medicine development. Four are described in the EMA announcement: a human-centric approach, a risk-based approach, adherence to standards and a clear context of use. The principles are non-binding.

For devices, the FDA keeps an official list of AI-enabled medical devices authorized for sale in the United States. The list is updated periodically and is not exhaustive. The FDA also says it will study how to identify and tag devices that use foundation models, including large language models, so tracking of generative AI in devices is not consolidated.

Quick Self-Check

Imagine a clinic tells you that a genetic test or an algorithm will guide your treatment. What would you ask first? As an editorial suggestion, not an official checklist, you could ask what the test or tool is meant to predict, whether the evidence comes from real patients or only from lab data, and who reviews the result with you.

Limits Worth Knowing

A 2025 methodological review hosted on PMC (Okyere and colleagues, Computational and Structural Biotechnology Journal) adds a wider warning, and it is a journal article, not an NIH position. According to that review, random train and test splits can leak information, and models can learn specific compounds instead of biology, which inflates accuracy. The authors call for stricter validation, including external validation on clinical datasets.

The same review names data bias as a limit. Many health datasets underrepresent genetic, ethnic and socio-environmental diversity, so models may be less accurate for underrepresented groups and could widen health disparities.

What This Means for You

AI in personalized medicine is real as research and increasingly present in drug development paperwork, but the examples in patient care are still early. Ask your prescriber what evidence supports any test or tool, and rely on official labeling and your clinician for decisions about your health.

Frequently Asked Questions

What is personalized medicine?

The NHGRI describes it as using an individual’s genetic profile to guide decisions about prevention, diagnosis and treatment, including the choice of drug, dose and schedule.

Is AI-based drug response prediction used in clinics today?

The example above, PERCEPTION from the NCI, is a proof-of-concept study validated on retrospective data. The NCI also says precision medicine is not yet routine care for most patients.

Does the Nobel Prize in Chemistry mean AI has produced approved drugs?

No. The 2024 prize recognized work in structural biology, a different step from approving a medicine.

Does the FDA recommend a genetic test for every drug on its pharmacogenetic table?

No. The FDA says that listing a pair does not mean testing is recommended, unless the test is a companion diagnostic, and prescribers should rely on approved labeling.

This content is for informational purposes only and does not constitute personalized financial advice. Terms and conditions change, so confirm the details with the official provider before making a decision.

Last reviewed: 2026-10-18

Sources consulted:

https://www.genome.gov/genetics-glossary/Personalized-Medicine
https://www.cancer.gov/about-cancer/treatment/types/precision-medicine
https://www.genome.gov/genetics-glossary/Pharmacogenomics
https://www.fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations
https://www.cancer.gov/news-events/press-releases/2024/ai-tool-matches-cancer-drugs-to-patients
https://pmc.ncbi.nlm.nih.gov/articles/PMC12722026/
http://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices
https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development
https://www.ema.europa.eu/en/news/ema-fda-set-common-principles-ai-medicine-development-0
https://www.ema.europa.eu/en/news/reflection-paper-use-artificial-intelligence-lifecycle-medicines
https://www.nobelprize.org/prizes/chemistry/2024/press-release/