AI in Precision Medicine: Personalized Medicine & Cancer Care (2026)

Last updated: October 14, 2026

Artificial intelligence (AI) is helping medicine move beyond the traditional approach of treating patients according to an average profile. By analysing complex information from genetic testing, medical imaging, laboratory results, electronic health records and other sources, AI may help clinicians identify patterns that influence disease risk, diagnosis, prognosis and treatment selection.

This is the promise of AI in precision medicine: combining individual biological and clinical information with increasingly sophisticated computational tools to make healthcare decisions more specific to the person receiving care.

The potential is particularly important in oncology. Two people with cancer in the same organ may have different mutations, molecular pathways, immune environments and treatment responses. AI can help researchers and clinical teams interpret these differences, investigate possible treatment options and identify relevant clinical trials.

However, an important distinction must be maintained. A computational model that predicts an outcome in a research dataset has not necessarily demonstrated that using the model improves outcomes for real patients. Clinical validation, appropriate regulatory authorization, evidence of benefit and professional oversight remain essential.

This guide examines how AI is being used in precision medicine, what the evidence supports, which applications remain experimental, and how patients and healthcare professionals can evaluate AI-assisted medical decisions.

Key takeaway: Precision medicine defines the goal of tailoring care to the individual. AI is one set of tools that may help achieve that goal. It does not replace validated diagnostic tests, clinical evidence, qualified healthcare professionals or informed patient decisions.

In this guide

• What is precision medicine?
• How AI works in personalized healthcare
• Eight applications of AI in precision medicine
• AI in precision oncology and cancer treatment
• AI, biomarkers and molecular testing
• How strong is the evidence?
• Limitations, risks and ethical concerns
• Questions patients should ask
• The future of precision medicine
• Frequently asked questions
• Scientific references and further reading

1. What Is Precision Medicine?

Precision medicine, sometimes called personalized medicine, is an approach to prevention, diagnosis and treatment that accounts for differences between individuals. These differences may include genetic makeup, molecular biomarkers, medical history, disease characteristics, environmental exposures, lifestyle and other clinically relevant factors.

The US Food and Drug Administration (FDA) describes precision medicine as an approach that considers individual differences, including genes, environment and lifestyle, when tailoring care. It also emphasizes the role of accurate diagnostic testing in determining which treatments may be appropriate.

Source: FDA: Precision Medicine.

Traditional medicine versus precision medicine

Dimension	Traditional approach	Precision medicine approach
Patient assessment	Symptoms, examination and standard clinical tests	Standard assessment plus relevant individual biological and clinical characteristics
Disease classification	Primarily based on clinical presentation and established diagnostic categories	May incorporate molecular subtypes, biomarkers and other disease-specific features
Treatment selection	Evidence-based standard treatments for the patient's diagnosis	Evidence-based treatments selected using additional individual information when clinically useful
Risk assessment	Validated population and clinical risk factors	May combine validated risk factors with genetic, molecular or longitudinal data
Monitoring	Symptoms, examination, laboratory tests and imaging	Those established methods, potentially supplemented by validated digital monitoring tools

These approaches are complementary rather than mutually exclusive. Precision medicine still depends on conventional clinical evaluation, established treatment protocols and the results of appropriately validated tests.

Where does AI fit?

AI can analyse patterns in datasets too large or complex for straightforward manual assessment. Depending on the application, it may help to:

• Recognize patterns in medical images, pathology slides and molecular data.
• Identify relationships between biomarkers and disease characteristics.
• Estimate an individual's risk of a particular outcome.
• Organize medical information for clinician review.
• Identify potentially relevant treatments or clinical trials for further evaluation.
• Generate research hypotheses that can be tested experimentally.

These capabilities differ substantially in their clinical readiness. Image analysis tools with an established diagnostic role should not be treated as equivalent to experimental AI models that predict how a cancer will respond to a drug.

2. How AI Works in Precision Medicine

AI is an umbrella term covering several computational methods. Understanding the differences helps clarify what a particular medical tool can and cannot do.

Machine learning

Machine learning (ML) algorithms learn statistical relationships from data. A model may be trained on patient records containing clinical characteristics and known outcomes, then evaluated on separate data to determine whether it can make useful predictions for new patients.

Applications include risk prediction, classification of disease subtypes and estimation of treatment outcomes. A model's usefulness depends on the quality of its training data, its validation and whether its predictions improve a real clinical decision.

Deep learning

Deep learning uses multilayer neural networks to identify patterns in complex data. It has become important in medical imaging, digital pathology, molecular biology and other tasks involving high-dimensional information.

For example, an image-analysis model may identify suspicious patterns on a scan or quantify features within a digitized tissue sample. The findings still need to be interpreted in the context of the test's intended use and the patient's clinical circumstances.

Natural language processing and large language models

Natural language processing helps computers interpret written or spoken language. In healthcare, it can support the extraction of information from medical records, clinical research literature, pathology reports and treatment guidelines.

Large language models and multimodal AI systems can also help organize information from several sources. However, generative systems may produce inaccurate statements, omit important context or invent references. Their outputs require verification, particularly when they influence medical decisions.

The World Health Organization has published specific guidance on the ethics and governance of large multimodal models in healthcare, reflecting the need to assess safety, transparency, accountability and effects on health equity.

Source: WHO: Ethics and Governance of AI for Health — Guidance on Large Multimodal Models.

Multimodal AI

Multimodal AI combines different types of information, such as genetic variants, pathology images, radiology scans, laboratory results and clinical notes. The goal is to identify relationships that might be missed when each data source is examined separately.

In precision oncology, this is particularly relevant because cancer behaviour reflects multiple interacting factors. Nevertheless, combining more data does not automatically produce a more accurate or useful model. Every input must be reliable, relevant and appropriate for the question being investigated.

3. Eight Applications of AI in Precision Medicine

1. Genomics and genetic variant interpretation

Genomic testing can identify large numbers of genetic variants. Only some have established clinical significance, and their interpretation may depend on the disease, the specific variant, the available evidence and the context in which the result is used.

AI and other computational methods can help prioritize variants, identify patterns in sequencing data and connect findings with published scientific evidence. They may also help researchers examine relationships between genetic changes and disease mechanisms.

In clinical practice, however, an AI-generated interpretation must not be confused with a definitive diagnosis. The significance of a variant should be assessed using appropriate laboratory methods, established classification standards, disease-specific evidence and qualified clinical interpretation.

AI can also contribute to the investigation of rare diseases, where a patient's clinical features and genetic findings may be difficult to reconcile. The extent to which a particular tool improves diagnosis must be established for its intended population and use.

2. Precision oncology and treatment selection

Cancer is not one disease. Tumours can differ in their molecular features, tissue characteristics, immune environments and mechanisms of resistance, even when they originate in the same organ.

AI may support precision oncology by integrating molecular profiling, pathology images, imaging studies and clinical records. Potential uses include identifying disease subtypes, prioritizing biomarkers for review, estimating outcomes and finding clinical trials for which a patient might be eligible.

For example, a patient with a specific tumour alteration may be considered for a targeted treatment if that alteration and the relevant clinical circumstances meet established treatment criteria. AI may help organize the evidence, but the molecular result, treatment indication and current clinical guidance remain central.

Further discussion: National Cancer Institute: Biomarker Testing for Cancer Treatment.

3. Digital pathology and medical imaging

AI-based image-analysis systems can examine radiology studies and digitized pathology slides, helping clinicians identify, classify or quantify features of interest.

Depending on the specific system, applications may include:

• Highlighting suspicious regions on medical images.
• Supporting the measurement of tumour size or other imaging features.
• Helping pathologists assess tissue patterns.
• Quantifying features that might be associated with prognosis or treatment response.
• Supporting workflow prioritization and quality control.

Some AI-enabled medical devices have received FDA marketing authorization for defined clinical uses. However, the existence of an authorized imaging tool does not establish that every AI system can reliably diagnose cancer, determine its molecular profile or predict treatment response.

Source: FDA: Artificial Intelligence-Enabled Medical Devices.

4. Pharmacogenomics and medication selection

Pharmacogenomics examines how genetic differences may influence a person's response to medicines. Some genetic variants affect drug metabolism, exposure, efficacy or the risk of adverse reactions.

AI may help researchers discover relationships between genetic features and drug response, interpret complex medication histories or combine pharmacogenomic information with clinical variables. These applications must be distinguished from established gene-drug recommendations.

Where a medicine has a validated pharmacogenomic recommendation, clinicians should refer to authoritative guidance and the applicable drug labeling. An AI system should not override established dosing instructions or make unsupported recommendations based on speculative gene-drug associations.

Useful resources include the Clinical Pharmacogenetics Implementation Consortium (CPIC) and the FDA's Table of Pharmacogenetic Associations.

5. Biomarker discovery and multi-omics

Precision medicine increasingly uses different kinds of biological measurements, sometimes collectively called multi-omics. These may include genomics, transcriptomics, proteomics and metabolomics, alongside clinical and environmental information.

Machine learning can identify patterns within these datasets and help researchers propose potential biomarkers for diagnostic, prognostic or treatment-related use.

For example, a research model might identify a molecular signature associated with a higher probability of treatment response in a particular patient group. That finding is a starting point for validation, not proof that the signature can already guide routine treatment.

Before a new biomarker is used clinically, researchers must establish its analytical reliability, its clinical validity for the intended purpose and, where applicable, whether using it improves patient management or outcomes.

6. Predicting treatment response and adverse effects

Two patients with similar diagnoses may experience different outcomes with the same treatment. AI models can be trained to investigate whether characteristics such as age, comorbidities, laboratory results, molecular features and prior treatments are associated with different outcomes.

Potential uses include predicting the probability of complications, identifying patients who require closer monitoring and estimating the likelihood of response to a treatment.

The important limitation is that a statistical association does not necessarily show that a treatment caused an outcome or that selecting treatment using the model will improve results. A clinically useful prediction tool must demonstrate suitable accuracy, calibration and benefit in the intended population.

7. Clinical trial matching

Clinical trial eligibility criteria can be complicated. Studies may specify particular diagnoses, biomarker results, previous treatments, laboratory thresholds and other inclusion or exclusion criteria.

AI-assisted systems may search trial databases, extract eligibility requirements from study descriptions and compare them with information in a patient's medical record. This can help patients and clinical teams find studies worth investigating.

However, an apparent match is not confirmation of eligibility or enrolment. Trial status can change, records may be incomplete, and eligibility often requires detailed review by the trial team.

Patients can also search the official ClinicalTrials.gov database and discuss relevant studies with their treating clinicians.

8. Personalized prevention and longitudinal monitoring

AI may help combine information from medical history, laboratory tests, validated risk models and selected wearable or digital-health measurements to identify patterns over time.

Potential applications include monitoring selected physiological signals, supporting diabetes management, estimating certain disease risks and identifying changes that warrant clinical review.

These applications are not interchangeable. Evidence that a device can measure a physiological signal accurately does not necessarily prove that it can diagnose a disease, predict a serious outcome or improve long-term health.

For preventive care, AI predictions should complement established screening recommendations and assessment of individual risk factors rather than replace them.

4. AI in Precision Oncology: From Tumour Profiling to Treatment

Precision oncology is one of the most important areas in which AI and personalized medicine intersect. Cancer care generates multiple types of data, and interpreting them together can be demanding for patients, clinicians and multidisciplinary molecular tumour boards.

A 2025 review in Nature discussed how AI may contribute across cancer prevention and diagnosis, treatment planning and monitoring, clinical trial matching, biomarker development and the discovery of potential new therapies. A 2026 perspective in Nature Reviews Cancer also examined the growing convergence of machine learning and genomics in precision oncology.

References:

• Hallmarks of artificial intelligence contributions to precision oncology, 2025.
• Convergence of machine learning and genomics for precision oncology, 2026.

The AI-assisted precision oncology workflow

Stage	Information examined	Potential contribution of AI	Important safeguard
1. Diagnosis	Symptoms, imaging, pathology and clinical history	Support detection, image analysis or classification	Diagnostic confirmation and clinical interpretation
2. Molecular profiling	Tumour DNA, RNA and other selected biomarkers	Help prioritize and interpret complex molecular findings	Reliable testing and clinically appropriate variant interpretation
3. Treatment assessment	Biomarkers, disease subtype, previous treatment and clinical condition	Organize evidence relevant to possible treatment options	Confirm the treatment's evidence, indication and suitability
4. Clinical trial search	Diagnosis, biomarkers, previous treatments and eligibility criteria	Identify potentially relevant research studies	Verify eligibility and trial status with the research team
5. Monitoring	Symptoms, laboratory results, imaging and other validated measures	Help identify changes and trends	Confirm clinically significant findings and appropriate follow-up
6. Resistance assessment	Progression, repeat testing and changes in tumour characteristics	Support research into possible resistance mechanisms	Do not infer an actionable resistance mechanism without adequate evidence

Why tumour heterogeneity matters

A tumour may contain multiple cell populations with different molecular characteristics. In addition, its biology can change over time, particularly under the selective pressure of treatment.

A single biopsy may therefore fail to capture every clinically relevant feature of a cancer. Imaging, repeat biopsy, liquid biopsy and other tests may each provide different information, depending on the disease and the clinical question.

AI research seeks to integrate these different sources of information and investigate how they relate to disease behaviour. This is promising, but models must account for sampling limitations, changes over time and differences between patients.

Can AI tell which cancer drug will work best?

Not reliably in every situation. Some cancer treatments are already selected using validated molecular tests and established treatment indications. Other AI-based predictions of drug response remain investigational.

A model might predict that a particular drug is associated with better outcomes among people whose tumours share a certain feature. To establish whether that prediction can guide treatment, researchers must determine whether it generalizes to the intended population and whether acting on the prediction improves decisions or patient outcomes.

Clinical rule: Do not select, stop, delay or substitute an established cancer treatment solely because an AI model, chatbot or commercial report recommends an alternative. Discuss the findings with the treating oncology team and, where relevant, a molecular tumour board.

Related SmartCancer content: Link this section to the Cancer Academy, Biomarker Guide, Immunotherapy Guide and Clinical Trial Matching Guide when the relevant pages are available.

5. AI, Biomarkers and Molecular Testing

A biomarker is a measurable biological characteristic that can provide information about a biological process, disease or response to an intervention. Biomarkers can be measured in tissue, blood and other biological samples, or sometimes through imaging and physiological measurements.

In oncology, biomarkers may help establish a diagnosis, define a tumour subtype, estimate prognosis, identify a treatment option or monitor disease. Their uses are specific to the biomarker, the test and the clinical question.

Four categories to distinguish

Category	Question answered	Example of appropriate interpretation
Diagnostic biomarker	Does the finding help identify or classify a disease?	Interpret the result using the validated test and diagnostic criteria.
Prognostic biomarker	Is the finding associated with the likely course of disease?	An association with outcome does not automatically identify a better treatment.
Predictive biomarker	Does the finding help identify patients more or less likely to benefit from a particular intervention?	Use the evidence relevant to the treatment, test and patient population.
Monitoring biomarker	Can changes in the finding help assess disease status or response?	Interpret changes alongside symptoms, examination and other relevant tests.

AI prediction versus a validated biomarker

AI can discover a feature that appears associated with treatment response. That feature may later become a candidate biomarker, but additional evidence is required before it can be used as a reliable clinical decision tool.

The FDA's resources on biomarkers and companion diagnostics explain how biomarkers and diagnostic tests contribute to medical-product development and treatment decisions.

A companion diagnostic is a test that provides information essential to the safe and effective use of a corresponding drug or biological product. Depending on the indication, it may identify patients likely to benefit, those at increased risk of serious adverse effects or patients who need monitoring to guide treatment.

The important lesson is that a biomarker's clinical meaning depends on its intended use. Finding a mutation, generating an AI score and establishing eligibility for an approved treatment are three different things.

Why a negative result needs context

A negative test result does not always prove that a tumour lacks every potentially relevant molecular alteration. Results can depend on the specimen, tumour content, assay coverage, detection limits and the biology of the disease.

For example, the FDA-approved labeling for some circulating tumour DNA tests explains that a negative plasma result may need follow-up with tumour tissue testing when clinically appropriate and feasible.

Patients should ask what the test assessed, which alterations it can detect, whether the sample was adequate and whether additional testing is appropriate in their clinical situation.

Source: FDA: FoundationOne Liquid CDx approval information.

6. AI in Other Areas of Personalized Healthcare

Although precision oncology is an important example, the broader concept applies to many areas of medicine.

Rare and genetic diseases

AI can help clinicians and researchers compare a patient's clinical features with genomic findings and established disease knowledge. This may support prioritization of diagnostic possibilities, particularly when symptoms involve multiple organ systems.

The resulting hypotheses still require appropriate clinical, genetic and laboratory assessment. A model's confidence score is not equivalent to a confirmed diagnosis.

Cardiovascular medicine

AI may help interpret electrocardiograms and cardiac imaging, or combine selected clinical measurements to assess risk. Whether an application is suitable for routine care depends on its intended use, validation and evidence that acting on its output benefits patients.

Metabolic health and diabetes

Personalized metabolic care may combine laboratory results, medical history, glucose measurements, medications and lifestyle factors. Some digital systems use algorithms to support monitoring or medication management, including insulin dosing in appropriately authorized systems.

AI-generated dietary, supplement or medication recommendations require particular care. Algorithms may not fully account for kidney function, drug interactions, nutritional deficiencies, pregnancy, comorbidities or the quality of the underlying evidence.

Drug development

AI is also used in biomedical research to prioritize molecular targets, search chemical libraries, predict molecular properties and propose drug candidates. These methods can help researchers decide which hypotheses to test experimentally.

Discovering a promising molecule is not the same as developing a safe and effective medicine. Experimental studies, clinical trials and regulatory assessment remain necessary to establish whether a candidate benefits patients.

7. How Strong Is the Evidence for AI in Precision Medicine?

Evidence should be assessed against the exact claim a tool makes. A system designed to flag suspicious images has a different purpose from one that predicts survival or recommends a particular cancer treatment.

A 2025 scoping review in BMC Medical Research Methodology examined machine-learning models for oncology decision-making that had external validation and clinical-utility assessments. The review included 56 studies published from 2018 through 2022 and highlighted ongoing issues involving generalizability, validation across populations, calibration and consistency in reporting.

The review also illustrates why promising technical performance should not automatically be interpreted as evidence of improved patient outcomes.

Source: Externally validated and clinically useful machine learning algorithms to support patient-related decision-making in oncology: a scoping review.

A practical evidence ladder for reviewing AI tools

The following E0–E5 framework is an editorial checklist for organizing evidence. It is not a universal regulatory classification or an official grading system for all medical AI tools.

Level	Evidence stage	What it establishes	What it does not establish
E0	Concept or hypothesis	A proposed use or plausible biological rationale	Accuracy, reproducibility or clinical usefulness
E1	Technical or preclinical evaluation	Initial performance on selected datasets or experimental systems	Reliable results in diverse clinical populations
E2	Retrospective clinical evaluation	Performance on historical patient data	Proof that results generalize to new patients or change clinical outcomes
E3	Independent external validation	Testing on separate data representative of the intended setting	Proof that using the tool improves care
E4	Prospective clinical evaluation	Assessment in an actual clinical workflow, potentially including effects on decisions	Definitive proof of improved patient outcomes in every setting
E5	Strong clinical utility evidence	Evidence appropriate to the claim, potentially including comparative or randomized trials showing patient-relevant benefit	Universal suitability for every patient, disease or healthcare system

Regulatory authorization is a separate consideration. An authorized medical device may have an established use without proving every broader claim a manufacturer or third party might make about AI and personalized healthcare.

What should a good validation study report?

• Intended use: The specific clinical task and population for which the model was designed.
• Independent validation: Performance on patients or data that were not used to develop the model.
• Calibration: Whether predicted probabilities correspond reasonably to observed outcomes.
• Discrimination: How well the model separates the outcomes it is intended to distinguish.
• Subgroup performance: Whether accuracy differs across relevant patient groups and clinical settings.
• Clinical utility: Whether using the model improves decisions, workflows or patient-relevant outcomes compared with an appropriate alternative.

A high accuracy score alone may be misleading, especially when an outcome is rare or a dataset differs from the population in which the tool will be used.

8. AI in Precision Medicine: Potential Benefits and Limitations

Potential benefit	Corresponding limitation
Faster analysis of complex datasets	Faster processing does not guarantee correct interpretation.
More consistent image analysis	Performance can vary across equipment, institutions and populations.
Identification of previously unrecognized patterns	Patterns may be spurious, biased or clinically irrelevant.
Support for biomarker interpretation	A candidate biomarker may lack sufficient clinical validation.
More efficient clinical trial searches	Eligibility and trial availability still require confirmation.
Integration of multiple patient data sources	Data may be incomplete, inconsistent or difficult to reconcile.
Support for individualized risk estimates	Predictions may be poorly calibrated for an individual or an unfamiliar population.

9. Risks, Bias and Ethical Concerns

Bias and unequal performance

AI systems learn from the data used to develop them. If certain populations, demographic groups or clinical settings are underrepresented, the resulting model may perform less reliably for those groups.

This matters in precision medicine because a recommendation that appears accurate overall may be less useful for a patient whose circumstances differ from the training data. Developers and healthcare institutions should assess performance across relevant subgroups and settings.

Data privacy and security

Precision medicine may involve sensitive information, including genetic data, diagnoses, pathology images and detailed medical histories. Such data can be useful for research, but collection, sharing and processing introduce privacy and security risks.

Patients should understand how their information will be used, who may access it, whether it will be shared with third parties and what protections apply. Healthcare organizations should follow applicable privacy requirements and appropriate data-governance practices.

Hallucinations and fabricated evidence

Generative AI can produce confident but incorrect statements, including inaccurate treatment summaries, misleading interpretations of laboratory results or references that do not support the claim being made.

Clinical teams and patients should verify material statements against primary research, authoritative guidelines, approved product information and validated test reports. An AI-generated bibliography should never be accepted without checking that the cited source exists and supports the claim.

Correlation is not causation

An AI model may find that a particular biomarker is associated with a favourable outcome among patients who received a treatment. This alone does not prove that the treatment caused the favourable outcome or that patients with the biomarker benefit more from that treatment than they would from another option.

Distinguishing these possibilities may require suitable comparative analyses, prospective studies and, where appropriate, randomized trials.

Overreliance on algorithms

AI should support rather than eliminate appropriate human judgment. Clinicians need to consider the patient's history, examination, preferences, treatment goals, comorbidities and the limitations of the data available.

The WHO's guidance on AI ethics emphasizes human autonomy, safety, transparency, accountability, inclusiveness and sustainability.

Source: WHO: Ethics and Governance of Artificial Intelligence for Health.

Regulatory authorization and changing software

AI-enabled medical products have different intended uses and regulatory pathways. In the United States, the FDA evaluates medical devices, including devices incorporating AI, under the applicable regulatory framework.

As of September 2026, the FDA's public information indicates that more than 1,600 AI-enabled medical devices have been authorized for marketing in the United States. This is a broad device category, not a count of precision-oncology treatment-selection systems, and authorization should not be interpreted as proof of every claimed clinical benefit.

Source: FDA: Artificial Intelligence-Enabled Medical Devices.

Authorization in one country also does not automatically establish authorization in another. Patients and clinicians should check the product's approved or cleared intended use, applicable local requirements and relevant evidence.

10. How Patients Can Evaluate an AI-Assisted Medical Recommendation

Patients do not need to understand the mathematics behind an AI model to ask useful questions about its reliability and role in care.

Patient checklist

• What clinical question is the AI tool intended to answer?
• Has the tool been independently validated in patients similar to me?
• Is it authorized or cleared for this specific intended use where required?
• Does it rely on a validated diagnostic test, or is it making a research-based prediction?
• Can the result be checked against published evidence and the original test report?
• Will the recommendation change my care, and is there evidence that this change improves outcomes?
• What are the alternatives, potential harms, uncertainties and costs?
• How will my health and genetic information be protected?

Additional questions for cancer patients

When AI is used to support cancer treatment decisions, patients may also ask:

• Which specific biomarker, molecular alteration or clinical feature supports the recommendation?
• Is the proposed treatment approved or recommended for this cancer type and clinical situation?
• Is the evidence based on clinical outcomes in patients, or primarily on laboratory experiments, retrospective data or computational predictions?
• Could a molecular tumour board or another oncology specialist review the result?
• Is there an appropriate clinical trial, and has the trial team confirmed eligibility?

Patients should not postpone time-sensitive treatment while pursuing experimental AI analyses or unvalidated treatment predictions. Decisions about changing care should be made with the treating team.

11. The Future of AI in Precision Medicine

Several developments could make precision medicine more informative and practical over time. Their clinical value will depend on rigorous evaluation rather than technical sophistication alone.

Multimodal foundation models

Newer AI systems can process multiple kinds of data, including text, images and selected molecular information. In principle, this could help clinicians and researchers interpret complex patient records more efficiently.

The challenge is to establish which tasks these systems can perform reliably, how their outputs should be checked and whether using them improves care in real clinical settings.

Longitudinal and real-world data

Repeated measurements can provide information about how a person's condition changes over time. Integrating reliable longitudinal data may improve research into disease progression, treatment response and risk prediction.

However, real-world data are not automatically high-quality evidence. Missing information, differences in clinical practice and selection bias can all affect what a model learns.

AI-supported drug discovery

AI is being used to identify biological targets, prioritize candidate compounds and explore potential treatment combinations. In oncology, these methods may help researchers investigate tumour vulnerabilities and mechanisms of resistance.

Nevertheless, computational predictions require experimental testing and clinical development. A molecule generated or selected by AI is not a proven treatment, and a predicted drug combination should not be interpreted as a recommendation for patients outside appropriate evidence-based care.

More transparent and clinically accountable systems

Responsible deployment will require clearer reporting of data sources, intended use, performance, limitations and subgroup results. Healthcare organizations will also need processes to detect changes in performance when patient populations, clinical workflows or software change.

The goal is not simply to build a more powerful model. It is to create a reliable clinical process in which the model's contribution can be measured and its mistakes can be identified and addressed.

12. What AI in Precision Medicine Means for OneDayMD and SmartCancer

A useful precision-medicine knowledge hub should connect four elements:

1. Entity: A disease, gene, genetic alteration, biomarker, treatment, diagnostic test or AI system.
2. Relationship: How two entities are related, such as a biomarker predicting eligibility for a particular treatment.
3. Evidence: The underlying study, validation data, clinical guideline, test authorization and limitations.
4. Decision: What the evidence supports in a defined clinical context, including uncertainty and alternatives.

For cancer, the key information architecture could connect:

Knowledge area	Questions the knowledge hub should answer
Cancer type and histology	Which disease and subtype were actually studied?
Gene or biomarker	What was measured, using which test, and with what analytical limitations?
Treatment	What therapy was investigated, and for which clinical indication?
Resistance and response	Was response measured in laboratory models, retrospective cohorts or prospective patient studies?
AI model	What task was the model trained to perform, and how was it validated?
Clinical evidence	Does the evidence demonstrate technical performance, clinical validity or improved patient outcomes?
Clinical decision	What can reasonably be concluded for this patient population, and what remains uncertain?

This structure would allow readers to move from a general AI topic to a specific biomarker or treatment and then examine the supporting evidence without implying that every association is clinically actionable.

It also offers a practical editorial distinction: established care, promising but incompletely validated research, and speculative claims should be labelled separately.

13. Frequently Asked Questions

What is AI in precision medicine?

AI in precision medicine refers to using computational methods to analyse individual biological and clinical information to support diagnosis, risk assessment, biomarker interpretation, treatment planning, monitoring or medical research. Its value depends on the intended use and evidence supporting the specific application.

Can AI personalize cancer treatment?

AI can support some aspects of personalized cancer care, including image analysis, genomic interpretation, research into treatment response and clinical trial matching. The evidence varies by application. Treatment decisions should be based on validated tests, relevant clinical evidence and professional assessment rather than an unverified AI recommendation.

Can AI identify the best cancer drug for an individual patient?

Not reliably in every situation. Some treatment decisions can already be guided by validated biomarkers and established indications. Broader AI-based drug-response predictions remain investigational in many settings and require independent validation and evidence of clinical usefulness.

What is the difference between AI and precision medicine?

Precision medicine is an approach to tailoring healthcare decisions to individual characteristics. AI is a family of computational technologies that can help analyse relevant information. Precision medicine can be practised without AI, and AI can be used for purposes that do not involve precision medicine.

What is the difference between a biomarker and an AI prediction?

A biomarker is a measurable biological characteristic with a defined interpretation in a particular context. An AI prediction is an output from a model. AI may help discover or interpret candidate biomarkers, but a model-generated association does not automatically establish a clinically validated biomarker.

Can AI replace a doctor in precision medicine?

AI can support specific tasks, but it does not replace the full clinical assessment required for individualized care. Appropriate treatment decisions involve the patient's condition, medical history, test reliability, evidence, preferences, risks and available alternatives.

Is an FDA-authorized AI medical device guaranteed to work for every patient?

No. Authorization relates to the applicable regulatory requirements and the device's intended use. Performance and suitability can still depend on patient characteristics, the clinical setting and whether the device is used as intended. Authorization is not a guarantee of benefit for every patient.

Can AI predict which treatment will cause fewer side effects?

Some models investigate the likelihood of adverse effects or complications. Whether a prediction is reliable enough to guide an individual treatment decision depends on its validation, intended use and supporting clinical evidence. A prediction should not replace established safety monitoring or professional advice.

How can patients protect their privacy when using medical AI?

Patients should review the service's privacy policy, understand how health and genetic information will be processed, and avoid submitting identifiable medical records to services without suitable privacy safeguards. Healthcare providers should follow the applicable requirements for protecting sensitive health information.

What is the future of AI in precision medicine?

Important research directions include multimodal data integration, improved genomic interpretation, digital pathology, personalized risk assessment, drug development and clinical trial matching. Progress will depend on independent validation, prospective evaluation, patient safety, privacy, equitable performance and evidence that using AI improves clinically meaningful outcomes.

14. Scientific References and Further Reading

The following resources provide background on precision medicine, biomarkers, medical AI, clinical validation and responsible implementation.

1. US Food and Drug Administration. Precision Medicine. An overview of precision medicine, genomic testing and the role of reliable diagnostics.
2. National Cancer Institute. Biomarker Testing for Cancer Treatment. Explains how cancer biomarker tests can help inform treatment decisions.
3. US Food and Drug Administration. Artificial Intelligence-Enabled Medical Devices. Regulatory background and the FDA's public resource on AI-enabled medical devices.
4. World Health Organization. (2024; published online March 2025 listing). Ethics and Governance of Artificial Intelligence for Health: Guidance on Large Multimodal Models. Guidance on the responsible use of multimodal AI systems in healthcare.
5. Reardon B, Culhane AC, Van Allen EM. (2026). Convergence of Machine Learning and Genomics for Precision Oncology. Nature Reviews Cancer. A perspective on integrating genomic data and machine learning in precision oncology.
6. Hallmarks of Artificial Intelligence Contributions to Precision Oncology. (2025). Full text on PubMed Central. Reviews potential AI applications across cancer diagnosis, treatment optimization and drug discovery.
7. Sousa Santos C, Amorim-Lopes M. (2025). Externally Validated and Clinically Useful Machine Learning Algorithms to Support Patient-Related Decision-Making in Oncology: A Scoping Review. BMC Medical Research Methodology. Examines external validation and clinical utility in oncology decision support.
8. O'Connor O, McVeigh TP. (2025). Increasing Use of Artificial Intelligence in Genomic Medicine for Cancer Care — The Promise and Potential Pitfalls. BJC Reports. Reviews opportunities and limitations of AI in cancer genomic medicine.
9. National Cancer Institute. Cancer Diagnosis Research. Background on diagnostic technologies, molecular alterations and treatment-related biomarkers.
10. World Health Organization. Ethics and Governance of Artificial Intelligence for Health. Foundational guidance on safety, accountability, transparency and equity.

Conclusion: AI Should Make Precision Medicine More Evidence-Based

AI has the potential to help clinicians and researchers interpret the increasingly complex information generated by modern medicine. Its applications range from image analysis and genomic interpretation to biomarker discovery, treatment-response research and clinical trial matching.

The opportunity is substantial, particularly in cancer care, where disease heterogeneity and the volume of molecular information make treatment decisions challenging.

But precision medicine should not become a label for any algorithm that produces a personalized-looking answer. The most important questions remain whether a tool is accurate for its intended purpose, whether it has been independently validated, whether the recommendation is supported by reliable evidence, and whether using it improves care.

The goal is not simply to make medicine more personalized. It is to make personalized medicine safer, more accurate, more equitable and more useful to patients.

Medical disclaimer: This article is for general educational purposes and is not a substitute for individualized medical advice, diagnosis or treatment. AI outputs and research findings should not be used independently to start, stop or change medication or cancer treatment. Discuss relevant decisions with a qualified healthcare professional.

Editorial approach: independent evidence review. Distinguish established clinical applications from emerging research and unvalidated hypotheses. No AI-generated treatment prediction should be represented as a proven clinical benefit without evidence supporting that claim.