Allergy, Immunology, and Medical Genetics

Medical genetics generates thousands of variants per patient from exome and genome sequencing. Determining which variants cause disease requires integration of population frequency, functional evidence, literature, inheritance, and clinical phenotype. Computational systems can retrieve evidence, apply criteria, and prioritize variants, but a ranking or pathogenicity score does not replace laboratory classification, clinical correlation, or confirmatory testing. Facial-phenotyping systems can rank syndromic hypotheses, while pharmacogenomic rules can deliver maintained prescribing guidance at the point of care. These are different technologies with different evidence requirements.

Learning Objectives

After reading this chapter, readers will be able to:

  • Evaluate AI systems for genetic variant classification and interpretation
  • Understand AI applications in allergic disease phenotyping
  • Assess immunodeficiency diagnostic support tools
  • Navigate the role of AI in precision medicine and pharmacogenomics
  • Recognize the unique challenges of rare disease AI
  • Apply evidence-based frameworks for genetics and immunology AI

The Clinical Context: These specialties deal with complex pattern recognition (allergic phenotypes, immune dysregulation, genetic variant interpretation) that AI may assist with, but face challenges of rare conditions, limited training data, and evolving knowledge bases.

What Works Well:

Application Evidence Level Key Benefit
Genetic variant evidence support Established decision support More consistent retrieval and application of specified criteria
Pharmacogenomics CDS (CPIC guidelines) Established non-AI comparator Evidence-based genotype-guided prescribing rules
Facial dysmorphology AI (Face2Gene) Moderate Rare syndrome recognition support

What’s Emerging:

Application Status Notes
Polygenic risk scores Variable Disease-, population-, and pathway-specific implementation evidence
Allergic phenotyping Research Asthma endotypes, food allergy prediction
Primary immunodeficiency classification Research Pattern recognition from clinical data

Critical Challenges:

  • Population bias: Most variant databases overrepresent European ancestry
  • Novel variants: AI cannot classify variants not in training data
  • Rare diseases: Limited training data by definition
  • Phenotype quality: AI performance depends on accurate clinical description

The Bottom Line: Computational variant tools can make evidence review more reproducible, but their results remain tool- and dataset-specific. CPIC pharmacogenomic decision support is a mature rules-based comparator rather than proof of machine-learning efficacy. Facial phenotyping can rank hypotheses in selected settings, and allergy and immunology AI remains largely research-stage. Polygenic risk scores require population-matched validation, calibration, an actionable clinical pathway, and comparative evidence before outcome-benefit claims.

Introduction

A single exome sequence generates thousands of variants, and determining which one contributes to disease requires integrating population frequency, functional evidence, literature, phenotype, and inheritance. Computational tools can automate parts of evidence collection and prioritization, but variants of uncertain significance remain a clinical and communication challenge. In a cohort of 3,272,035 people tested for hereditary disease, 53.6% of initially reported variants were classified as VUS; 4.6% of observed variants were reclassified over the study period, and most VUS reclassifications moved toward benign or likely benign categories (Kobayashi et al., 2024). Uncertainty is resolved by new evidence and structured reinterpretation, not by relabeling a model score as certainty.

Part 1: Genetic Variant Interpretation AI

The Classification Challenge

Exome and genome sequencing identify thousands of variants per patient. Determining which variant(s) cause disease requires integrating:

  • Population frequency (rare variants more likely pathogenic)
  • Functional impact predictions (protein truncation, missense effects)
  • Published literature and case reports
  • Clinical phenotype match
  • Segregation in family members

The ACMG/AMP guidelines (Richards et al., 2015) standardized variant classification into five categories: pathogenic, likely pathogenic, variant of uncertain significance (VUS), likely benign, and benign. Rules engines and machine-learning systems can support evidence retrieval, criteria application, and variant prioritization. They do not independently establish clinical validity.

AI-Assisted Classification Tools

Examples of computational support tools:

Tool Approach Evidence
InterVar Rules-based ACMG/AMP criteria application Candidate classification support
Franklin (Genoox) Evidence aggregation and interpretation workspace Commercial platform; verify current claims and local validation
Illumina Emedgene Phenotype-driven variant prioritization Commercial platform; registration or listing is not authorization for autonomous diagnosis
VarSome Evidence aggregation and ACMG/AMP support Classification support with maintained knowledge sources

Performance: Nicora and colleagues tested one penalized-logistic-regression approach that combined ACMG/AMP criteria with annotation features and reported that it resolved more VUS cases than the comparison rules-based and in silico tools on the evaluated datasets (Nicora et al., 2022). The study does not support a universal concordance percentage for automated ACMG classification. Every tool requires validation on the variant classes, genes, populations, and workflow in which its output will be used.

In silico predictors:

Pathogenicity prediction algorithms (CADD, REVEL, AlphaMissense) provide scores that inform the PP3/BP4 ACMG criteria (computational evidence):

  • AlphaMissense (DeepMind, 2023): Trained on evolutionary conservation and protein structure
  • REVEL: Ensemble of 13 prediction algorithms
  • CADD: Integrates diverse annotations

Limitation: Computational predictors lack clinical validation for standalone classification. They support, not replace, comprehensive variant analysis.

ACMG/AMP Position on AI

ACMG Guidelines and AI-Assisted Classification

The foundational ACMG/AMP variant classification guidelines (Richards et al., 2015) established standards that AI tools now implement:

Core principles that apply to AI:

  • Classification requires integrating multiple evidence types
  • Population database frequency informs but does not determine pathogenicity
  • Functional studies provide strong evidence when available
  • Clinical correlation remains essential

Implications for AI tools:

  • AI-assisted classification should follow ACMG criteria, not proprietary algorithms
  • Novel variants not in training databases require caution
  • AI output requires review by certified molecular geneticists or pathologists
  • Patient consent should address AI-assisted analysis

Maintained criteria: Gene- and disease-specific specifications refine criteria application. Computational tools must document which guideline version and ClinGen specifications they implement. The UK Association for Clinical Genomic Science published 2024 practice guidelines that address use of computational evidence within an expert interpretation process.

Population Bias in Variant Databases

A critical equity issue is that genomic resources and study cohorts remain unevenly representative across populations. The consequences depend on the database, gene, phenotype, population structure, and interpretation workflow.

Representation questions for an adoption review:

Resource feature Question
ClinVar submissions Which laboratories, evidence levels, and populations support the classification?
Population frequency resources Is coverage adequate for the relevant ancestry, geography, and variant class?
Development cohorts Does the cohort represent the patients in whom the system will be used?

Clinical consequence: Underrepresentation can make a population-specific benign variant appear unusually rare and can worsen uncertainty, calibration, or transportability. In the large Kobayashi cohort, VUS reclassification rates normalized to the number of people tested were higher in several underrepresented race, ethnicity, and ancestry groups, underscoring the need for careful reinterpretation rather than fixed database percentages (Kobayashi et al., 2024).

Mitigation: More diverse population resources can reduce gaps, but inclusion alone is insufficient. Laboratories should document ancestry handling, frequency thresholds, uncertainty, reanalysis policies, and how population data affect each classification.

Part 2: Facial Recognition for Genetic Syndromes

Face2Gene and DeepGestalt

Face2Gene uses the DeepGestalt deep convolutional neural network to identify syndromic patterns from facial photographs and generate a ranked differential diagnosis. The number of supported syndromes changes as the commercial platform is updated, so adoption review should rely on current documentation rather than a frozen catalog count.

Performance data:

Study Setting Reported Accuracy
Gurovich et al., 2019 Multi-center validation 91% top-10
Carrer et al., 2024 Italian pediatric clinic 98% correct in differential
Yahya et al., 2025 Genetic counseling unit 56% top-3 accuracy

Use case: Face2Gene is most useful for: - Generating differential diagnosis for unfamiliar dysmorphic features - Supporting non-geneticists in recognizing syndromic patterns - Ultra-rare conditions that non-specialists may not consider

Current deployment: Face2Gene is commercially available, but adoption figures reported by its manufacturer are not evidence of diagnostic accuracy, generalizability, or patient benefit.

Limitations

Ethnic and Racial Bias in Facial Recognition AI

Facial recognition AI for syndromes faces significant bias concerns:

Representation risk: Image provenance and subgroup composition are incompletely reported in many facial-phenotyping datasets. Performance can vary with: - African and African-American patients - East Asian patients - Middle Eastern patients - Mixed-ancestry patients

Clinical implication: Do not rely solely on Face2Gene for patients from underrepresented populations. Clinical gestalt and genetic testing remain essential.

Photo quality dependence: Performance varies with: - Lighting conditions - Image resolution - Patient age (some syndromes more recognizable in childhood) - Presence of glasses, facial hair, or other features

A 2024 retrospective comparison used 4,796 images covering 486 genetic syndromes plus 323 control images. DeepGestalt had 88% mean top-30 sensitivity across supported syndromes, while age and sex affected the separate D-Score dysmorphism output (Reiter et al., 2024). These selected-image results do not establish performance in an unselected referral population or safe use without genetics expertise.

Appropriate Use

Face2Gene should: - Generate hypotheses, not diagnoses - Supplement, not replace, clinical dysmorphology assessment - Be used with awareness of population limitations - Lead to confirmatory genetic testing, not standalone management changes

Part 3: Pharmacogenomics Clinical Decision Support

CPIC Guidelines: The Evidence Base

The Clinical Pharmacogenetics Implementation Consortium (CPIC) provides peer-reviewed, evidence-based guidelines for translating pharmacogenomic test results into prescribing recommendations. CPIC guidance and EHR delivery rules are important comparators for AI, but a maintained genotype-to-therapy rule is not necessarily a machine-learning system. Current gene-drug guidance should be checked through the CPIC guideline directory rather than a frozen count.

Maintained scope: - Published guideline and update inventory changes over time - Gene-drug pairs carry different evidence and recommendation levels - Implementation requires a validated genotype result, prescribing context, and current guideline version - Local CDS governance determines alert timing, overrides, and result provenance

Key gene-drug pairs with strong evidence:

Gene Drugs Recommendation
CYP2C19 Clopidogrel Poor metabolizers: alternative antiplatelet
CYP2D6 Codeine, tramadol Poor/ultrarapid metabolizers: alternative analgesic
HLA-B*57:01 Abacavir Positive: do not prescribe
TPMT/NUDT15 Azathioprine, 6-MP IM/PM: dose reduction
CYP2C9/VKORC1 Warfarin Genotype-based initial dosing

Pharmacogenomics CDS and AI

Electronic health record integration enables automated PGx alerts. These may be conventional rules, statistical models, or a combination, so evaluation should identify the actual computational component.

Implementation models:

  1. Pre-emptive testing: Genotype patients before first prescription; CDS fires when relevant drug ordered
  2. Reactive testing: Test when high-risk drug ordered; results guide dosing
  3. Hybrid: Pre-emptive panel for common genes; reactive for specialized drugs

Implementation evidence should report: - The exact guideline version and genotype-to-phenotype translation - Alert exposure, acceptance, override reasons, and prescribing changes - Clinical endpoints and adverse events when outcome-benefit claims are made - Whether the result reflects CPIC guidance, a local rule, or a learned model

AAAAI Framework for Augmented Intelligence

AAAAI Health Informatics, Technology, and Education Committee Framework

The American Academy of Allergy, Asthma and Immunology published a framework for AI in the specialty (Khoury et al., 2022):

Key recommendations:

  • Allergist-immunologists should be involved in design, validation, and implementation of AI tools for the specialty
  • Data science and bioinformatics training should be incorporated into fellowship programs
  • AI applications should be evaluated for bias before clinical deployment
  • Electronic health records provide rich data sources for AI in allergy

Identified applications:

  • Allergen component-resolved diagnostics
  • Asthma phenotype classification
  • Drug allergy de-labeling decision support
  • Immunodeficiency pattern recognition

Current status: Most cited applications remain research-stage. This work-group report is a specialty framework, not a formal validation standard for any product or an endorsement of a specific AI system.

Part 4: Allergic Disease AI Applications

Asthma Phenotyping

AI enables classification of asthma into clinically relevant subtypes:

T2-high vs. T2-low asthma: - T2-high: Eosinophilic, responds to biologics (omalizumab, mepolizumab, dupilumab) - T2-low: Neutrophilic or paucigranulocytic, limited biologic options

AI approaches: - Cluster analysis of clinical features, biomarkers, lung function - Treatment response prediction based on phenotype - Biomarker integration (FeNO, blood eosinophils, IgE)

Clinical status: Research-stage. Phenotyping tools not yet integrated into routine practice.

A 2025 European Respiratory Review systematic review and meta-analysis of 89 machine-learning trajectory studies found recurring childhood asthma, wheeze, and eczema patterns, including early-onset persistent, mid-onset persistent, and early-onset resolving trajectories. Most studies were rated low methodological quality, particularly for modeling and reporting, and few evaluated allergic multimorbidity, allergic rhinitis, or food allergy (Lisik et al., 2025). This supports trajectory modeling as useful epidemiology and risk stratification research, not a clinic-ready tool for biologic selection.

Food Allergy Prediction

AI models attempt to predict: - Which sensitized patients will react on oral food challenge - Severity of reactions - Development of tolerance

Inputs: - Skin prick test wheal size - Specific IgE levels and component testing - Clinical history - Age and other demographics

Performance: Published models use different populations, allergens, inputs, reference standards, and validation designs. A model should not replace an oral food challenge or specialist judgment unless its intended use and clinical pathway have been prospectively validated.

Drug Allergy Assessment

AI tools for: - Penicillin allergy de-labeling risk stratification - Cross-reactivity prediction (beta-lactam, sulfonamide families) - Distinguishing IgE-mediated from non-immune reactions

Clinical relevance: A 2025 rapid review reported that more than 90% of people with a penicillin-allergy label were not allergic after formal assessment (Ahmed and Sandoe, 2025). Risk stratification can identify candidates for an established delabeling pathway, but a model score is not itself a drug challenge or a confirmed delabeling result.

Part 5: Primary Immunodeficiency AI

The Diagnostic Challenge

Inborn errors of immunity comprise a growing and heterogeneous group of disorders. Many initially present through common infections, autoimmunity, inflammation, allergy, or malignancy, so the relevant diagnostic delay varies substantially by condition and health system. A fixed disease count or median delay becomes stale and should not be treated as a universal benchmark.

AI Approaches

Pattern recognition from clinical data: - Infection types, frequency, pathogens - Autoimmune features - Family history patterns - Growth and development

Genetic prioritization: - Phenotype-driven gene panel selection - Variant interpretation in immune-related genes - Immunologic phenotype-genotype correlation

Jeffrey Modell Foundation Warning Signs

AI could operationalize PID warning signs: - 4+ ear infections in one year - 2+ serious sinus infections in one year - 2+ months on antibiotics with little effect - 2+ pneumonias within one year - Failure to thrive - Recurrent deep skin or organ abscesses - Persistent thrush after age 1 - Need for IV antibiotics to clear infections - 2+ deep-seated infections - Family history of PID

Status: Research-stage. Pattern-recognition tools for inborn errors of immunity require external and prospective workflow validation before deployment claims.

Part 6: Polygenic Risk Scores

Current Evidence

Polygenic risk scores (PRS) aggregate weighted effects across many genetic variants to stratify risk relative to a reference population. Their meaning depends on the outcome, source population, score construction, calibration, and clinical pathway.

Cardiovascular disease: - Several scores have been evaluated in large cohorts - Integration studies test feasibility and communication as well as prediction - Incremental performance over established risk factors varies by score and population - Commercial availability does not establish clinical utility

Breast cancer: - Scores may refine risk estimates when combined with family history and established factors - Screening changes require evidence for the defined population and decision pathway - Enhancement and de-intensification raise different benefit-harm questions

Implementation status: The eMERGE Network selected and validated 10 PRS for implementation research in diverse US populations (Lennon et al., 2024). Implementation research does not establish a universal threshold or patient benefit for polygenic risk scores. A 2026 systematic review of 54 studies identified inconsistent reporting, limited guidelines, workflow barriers, and restricted generalizability as persistent translation problems (Martínez-Minguet et al., 2026).

Critical Limitations

Polygenic Risk Score Equity Concerns

Population transferability: - Most PRS developed in European populations - Predictive performance and calibration can decline when scores are transferred to ancestry groups underrepresented in development data - Calibration (absolute risk estimation) often poor outside development populations

Clinical utility uncertainty: - Does knowing PRS change patient behavior or outcomes? - How should PRS integrate with established risk factors? - Threshold for “high risk” classification varies by study

Current guidance: - PRS should not be used as standalone risk assessment - Population-matched validation required before clinical use - Genetic counseling should accompany PRS results - Equity review essential: PRS may widen rather than narrow health disparities

Part 7: Rare Disease Diagnosis

The Diagnostic Odyssey

  • Rare-disease evaluation often spans multiple specialties and repeated testing
  • Diagnostic delay differs by condition, age, phenotype, access, and health system
  • Many conditions are encountered first by clinicians unfamiliar with their complete presentation

AI Solutions

Phenotype-driven prioritization: - HPO (Human Phenotype Ontology) standardizes phenotype description - Exomiser, Phenomizer: Match patient phenotypes to disease databases - LIRICAL: Likelihood ratio-based phenotype matching

Facial analysis: - Face2Gene (discussed above) - GestaltMatcher: Open-source alternative

Multi-omic integration: - AI combines genomic, transcriptomic, methylation data - Helps identify variants when WES/WGS inconclusive

Undiagnosed Diseases Programs

The NIH Undiagnosed Diseases Program and Network use multidisciplinary evaluation that can include computational phenotype and variant analysis. Diagnostic yield depends on referral criteria, prior testing, cohort, and follow-up, so a fixed rate should not be treated as a program-wide AI effect.

AI contribution: Prioritizes variants, identifies candidate genes, suggests additional testing.

Part 8: Professional Society Positions

ACMG Position

The linked ACMG/AMP consensus guideline establishes sequence-variant interpretation criteria, not a product-specific AI endorsement. For computational tools, the defensible application is to document which criteria and evidence specifications are implemented, retain qualified laboratory review, and preserve clinical correlation. Broader claims about consent, model governance, or population performance require separate policies and evidence.

AAAAI Position

The AAAAI Health Informatics, Technology, and Education Committee published the 2022 augmented-intelligence work-group report discussed above. The Academy also issued a 2024 position statement recommending more precise EHR terminology for adverse reactions. The terminology statement addresses documentation and alerts; it is not a clinical AI validation guideline.

CPIC Implementation

CPIC provides maintained guidelines and implementation resources, including structured tables and standardized terminology that can support EHR CDS. The underlying recommendation and the software that delivers it should be evaluated separately.

Hypothetical Clinical Scenarios

The following fictional cases are decision exercises. They do not report actual patients, product outcomes, or legal standards.

Hypothetical case: A 3-year-old with developmental delay undergoes exome sequencing. A variant in a developmental-disorder gene is classified as VUS after an AI-assisted analysis. The parents ask whether it explains the child’s condition.

Question: How should the geneticist counsel about this VUS?

Discussion

Understanding VUS classification:

  • VUS means insufficient evidence to classify as pathogenic or benign
  • It does NOT mean “probably causes disease”
  • Many VUS are eventually reclassified as benign

Counseling approach:

  1. Explain that VUS cannot be used for clinical decision-making
  2. Describe factors that could lead to reclassification (more affected individuals, functional studies)
  3. Follow laboratory and clinical policies for reanalysis as evidence changes
  4. Consider additional family testing if feasible (segregation)
  5. Do not change management based on VUS

AI limitation: AI classified this as VUS because evidence is genuinely insufficient. The appropriate classification is uncertain, not “probably pathogenic.”

Teaching point: AI cannot resolve genuine uncertainty by confidence alone. Reclassification requires additional evidence, improved evidence interpretation, or both.

Hypothetical case: A pediatrician sees a 2-year-old with dysmorphic features and developmental delay. After obtaining appropriate consent and following institutional data-handling policy, the pediatrician uses a facial-phenotyping tool that lists Cornelia de Lange syndrome among its top three suggestions.

Question: What should the pediatrician do with this suggestion?

Discussion

Appropriate next steps:

  1. Review the syndrome features: Does the patient have cardinal features (synophrys, limb anomalies, growth restriction)?
  2. Refer to genetics: Face2Gene suggestions require expert evaluation
  3. Do not diagnose based on Face2Gene alone: The tool generates hypotheses, not diagnoses
  4. Genetic testing: If features are consistent, targeted gene panel or exome sequencing can confirm

Face2Gene value here:

  • Helped a non-specialist consider a rare diagnosis
  • Generated a hypothesis that would otherwise require dysmorphology expertise
  • Appropriate use: hypothesis generation leading to referral

What not to do:

  • Tell parents “the computer says your child has Cornelia de Lange syndrome”
  • Change management without genetic confirmation
  • Rely on Face2Gene for populations underrepresented in training data
Teaching point: Face2Gene is most valuable when it prompts consideration of diagnoses the clinician would not have considered. It should accelerate the path to genetics referral, not replace it.

Hypothetical case: A cardiologist orders clopidogrel for a 65-year-old patient after PCI. The EHR displays a pharmacogenomics alert: “Patient is a CYP2C19 poor metabolizer. Review current CPIC guidance and patient-specific alternatives.”

Question: How should the cardiologist respond?

Discussion

Clinical context:

  • CYP2C19 poor metabolizers have reduced conversion of clopidogrel to active metabolite
  • Higher rates of cardiovascular events on clopidogrel vs. normal metabolizers
  • CPIC recommends alternative antiplatelet (prasugrel or ticagrelor) for poor metabolizers

Appropriate response:

  1. Review the genotype result in the EHR
  2. Confirm patient has no contraindications to alternatives (bleeding risk, prior ICH)
  3. Apply current guidance together with indication, contraindications, bleeding risk, interactions, and local protocol
  4. Document the genotype result, source, and rationale for the prescribing decision

If cardiologist disagrees:

  • Document clinical reasoning for overriding alert
  • Consider that poor metabolizer status may have contributed to events in patients labeled “clopidogrel non-responders”

AI/CDS value:

  • Alert appeared at point of prescribing
  • Linked to evidence-based CPIC guideline
  • Actionable recommendation provided
Teaching point: Pharmacogenomics CDS is a mature example of rules-based precision-medicine support, not automatically an AI intervention. The current recommendation, genotype provenance, and patient context determine the appropriate action.

Hypothetical case: A 45-year-old woman asks about a direct-to-consumer genetic test that reports her coronary artery disease polygenic risk score in the 90th percentile. She is concerned about her heart disease risk.

Question: How should the physician counsel this patient?

Discussion

Understanding PRS results:

  • 90th percentile means her PRS is higher than 90% of the reference population
  • This does NOT mean 90% chance of heart disease
  • The associated relative and absolute risks depend on the specific score, outcome, reference population, and calibration

Counseling points:

  1. Contextualize: Ask about the reference population. If patient is non-European and score was developed in Europeans, interpretation is limited.

  2. Integrate with established factors: Review blood pressure, lipids, smoking, diabetes, family history, and a validated clinical risk pathway rather than interpreting the PRS in isolation.

  3. Actionability: What would she do differently? Standard cardiovascular prevention applies regardless of PRS.

  4. Limitations of DTC testing:

    • Validation varies by company
    • Quality of genetic counseling often inadequate
    • May not use most validated PRS algorithms

Recommended response:

  • Perform comprehensive cardiovascular risk assessment
  • Do not order additional testing based on PRS alone
  • Recommend standard primary prevention based on established guidelines
  • Note that elevated PRS may provide additional motivation for lifestyle modification
Teaching point: A DTC PRS should not drive a clinical decision without verification of the assay, score, reference population, calibration, and intended-use evidence. Established risk assessment and prevention guidance still applies.

Key Takeaways

Clinical Bottom Line

Genetic variant interpretation:

  • Computational tools can retrieve evidence and apply specified ACMG/AMP criteria, but performance is tool- and dataset-specific
  • VUS classification remains challenging; AI cannot resolve genuine uncertainty
  • Population bias in databases affects patients of non-European ancestry
  • All AI classifications require certified geneticist review

Facial recognition for syndromes:

  • A selected-image DeepGestalt experiment reported 91% top-10 accuracy; this does not establish unselected clinical performance
  • Useful for hypothesis generation, not diagnosis
  • Performance varies across ethnic groups
  • Most valuable for non-geneticists considering rare diagnoses

Pharmacogenomics CDS:

  • CPIC guidelines are the evidence-based standard
  • EHR-integrated CDS can deliver current PGx guidance at the point of prescribing
  • Pre-emptive genotyping enables point-of-care alerts
  • Current gene-drug coverage should be verified in CPIC’s maintained directory

Allergy and immunology AI:

  • Most applications remain research-stage
  • Asthma phenotyping and food allergy prediction are active research areas
  • AAAAI framework supports allergist involvement in AI development

Polygenic risk scores:

  • Clinical readiness varies by score, disease, population, calibration, and action pathway
  • Population transferability and equity are critical concerns
  • PRS should supplement, not replace, standard risk assessment

Rare disease diagnosis:

  • AI accelerates phenotype-to-gene matching
  • Facial analysis and HPO-based tools support diagnosis
  • Definitive diagnosis still requires genetic confirmation

Questions About AI in Allergy, Immunology, and Genetics

How accurate is AI for genetic variant classification?

Computational tools can standardize evidence retrieval and ACMG/AMP criteria application, but performance is tool- and dataset-specific. A variant of uncertain significance should not be converted into a clinical diagnosis or management decision solely from a model score.

What is Face2Gene and how accurate is it?

Face2Gene uses DeepGestalt to rank syndromic hypotheses from facial photographs. A retrospective 2019 experiment reported 91% top-10 accuracy on 502 selected images, but the output remains a differential aid that requires clinical and molecular confirmation.

Is there population bias in genetic variant databases?

Yes. Genomic resources and development cohorts remain unevenly representative. Ancestry, population structure, submission practices, and reference-dataset coverage can affect variant interpretation, VUS rates, and polygenic-score calibration.

Are polygenic risk scores ready for clinical use?

Some polygenic risk scores are being evaluated in clinical implementation studies, but utility depends on the disease, population, calibration, comparator, and actionable pathway. A percentile alone is not an absolute-risk estimate or a treatment recommendation.