HLA Assocation Studies

Allele Analysis

Predictive Value & Number Needed to Test

For a risk allele framed as a pre-prescription screening test, the positive predictive value (PPV) is the absolute risk of the reaction among exposed carriers, estimated from the allele’s sensitivity, the incidence among exposed individuals, and the population carriage frequency:

\[ \text{PPV} = \frac{s \times I}{f} \]

where \(s\) = sensitivity (proportion of cases carrying the allele), \(I\) = incidence of the reaction among exposed patients, and \(f\) = population carriage frequency.

Because s≤1s s≤1, PPV is bounded above by I/fI/f I/f, the ceiling reached when the allele is present in all cases.

The number needed to test (NNT) — number needed to screen to prevent one case, assuming the drug is withheld from carriers and avoidance is fully protective — is the reciprocal of the carrier-attributable risk:

\[ \text{NNT} = \frac{1}{s \times I} \]

Note

Both estimates are highly sensitive to the assumed incidence \(I\), usually the least precise input — report a range across plausible values. Low PPV is expected for HLA risk alleles with incomplete penetrance and does not by itself argue against utility; sensitivity and negative predictive value should be reported alongside, since an allele present in only a fraction of cases lacks the rule-out value of a near-necessary allele.

Inputs

Sensitivity (% of cases carrying allele)

Population carriage (%)

Incidence / prevalence (1 in N)

1 in 1,000 (0.10%)

Positive predictive value

—

of flagged carriers who react

Negative predictive value

—

Number needed to test

—

to prevent one case

— react — avoid the trigger unnecessarily per 1,000 flagged carriers

Ceiling at 100% sensitivity: incidence ÷ carriage = —. The test misses — of cases (non-carriers). NPV reads near 100% mainly when the phenotype is rare, not because the allele is necessary. Shaded band = typical SCAR incidence (1 in 1,000 to 1 in 10,000); more common HLA phenotypes (e.g., peanut allergy, ~1 in 50) sit to the left and yield higher PPV.

Power & Sample Size

Study

Cases (n)

Controls per case

Control carriage (%)

Evaluate OR

Significance

Family α

Tests (m) — Bonferroni

→ α = 0.05 per test

Target power (%)

Min detectable OR

—

at target power (exact)

Power at OR

—

Haplotype Analysis

Amino Acid Analysis

Shared-Peptide Binding Specificity Analysis

C1 vs C2

HLA-B -21 M/T Dimorphism

Bw4 vs Bw6

HLA Expression

HLA-A Expression

HLA-C Expression

BIGDAWG

^pappas2016?

https://github.com/IgDAWG/BIGDAWG

Hapl-o-Mat

^{schäfer2017?}

https://github.com/DKMS/Hapl-o-Mat

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-017-1692-y

Locus Analysis

midasHLA - Locus analysis

^migdal2021?

https://github.com/Genentech/midasHLA

Fine-Mapping

^clay2022?

Peptide Binding

MHCcluster2.0

^thomsen2013?

https://services.healthtech.dtu.dk/services/MHCcluster-2.0/

Docking Analysis

CB-Dock2

^liu2024?

^liu2022?

https://cadd.labshare.cn/cb-dock2/