Helix Insight

Documentation / Screening / Scoring Components

Scoring Components

Each variant is evaluated across seven independent dimensions. Every component produces a score between 0.0 (no signal) and 1.0 (strongest possible signal). The components are then combined using a weighted sum, where the weights depend on the screening mode and patient age group. See Screening Modes for weight profiles.

Component Overview

ComponentWhat It MeasuresKey Data Sources
ConstraintGene intolerance to variationpLI, LOEUF, missense Z-score
DeleteriousnessIn-silico pathogenicity predictionDANN, SIFT, AlphaMissense, MetaSVM, PhyloP, GERP
PhenotypePhenotype match or gene-disease burdenHPO overlap (diagnostic) or HPO count (screening)
DosageGene dosage sensitivityClinGen haploinsufficiency score
ConsequenceVariant consequence severityVEP consequence terms
Compound HetCompound heterozygote potentialSame-gene heterozygous variant pairs
Age RelevanceAge-appropriate gene prioritizationCurated gene lists per age group

1. Constraint Score

Measures how tolerant a gene is to variation using gnomAD constraint metrics. The scoring strategy adapts to the variant's consequence type, because the biological significance of constraint differs between loss-of-function and missense variants.

Loss-of-Function Variants

ConditionScore
pLI > 0.9 AND LOEUF < 0.351.0
pLI > 0.7 AND LOEUF < 0.50.8
pLI > 0.50.6
OtherwisepLI value directly

Missense Variants

ConditionScore
mis_z > 3.0 AND pLI > 0.90.9
mis_z > 2.00.7
pLI > 0.70.6
Otherwisemax(pLI, mis_z / 5.0)

2. Deleteriousness Score

Weighted aggregate of six in-silico pathogenicity predictors. Unlike ACMG classification which uses BayesDel as the single calibrated tool, the screening deleteriousness score combines multiple predictors for a broader assessment of variant impact. This is a ranking signal, not an ACMG evidence criterion.

PredictorWeightScoring
DANN40%Raw score (0.0-1.0)
SIFT15%1.0 if Deleterious, 0.0 if Tolerated
AlphaMissense15%1.0 if Pathogenic, 0.5 if Ambiguous, 0.0 if Benign
MetaSVM15%1.0 if Damaging, 0.0 if Tolerated
PhyloP7.5%Normalized: score / 10.0, capped at 1.0
GERP7.5%Normalized: score / 6.0, capped at 1.0

3. Phenotype Score

How the phenotype score is calculated depends on whether the patient has HPO terms.

ModeCalculation
Diagnostic (has HPO terms)Overlap between patient HPO terms and gene HPO associations, divided by total patient terms.
Screening (no HPO terms)Gene-disease burden: HPO count / 10, capped at 1.0. Genes associated with more phenotypes rank higher.

4. Dosage Score

Based on ClinGen haploinsufficiency curation, applied only to loss-of-function variants. A gene with sufficient evidence for haploinsufficiency means that losing one copy causes disease.

HI ScoreEvidence LevelDosage Score
3Sufficient evidence1.0
2Emerging evidence0.7
1Little evidence0.4
Missing or non-LoFNot applicable0.0

5. Consequence Score

Severity hierarchy based on the variant's predicted effect on the protein. More disruptive consequences receive higher scores.

ConsequenceScore
stop_gained, frameshift, splice_donor, splice_acceptor1.0
start_lost, stop_lost0.9
inframe insertion/deletion0.8
missense0.7
splice_region0.6
UTR (5' or 3')0.4
synonymous0.2
intronic0.1

6. Compound Heterozygote Score

Detects potential compound heterozygotes for autosomal recessive conditions. When two different heterozygous variants occur in the same gene, the patient may have both copies of the gene affected -- one variant on each chromosome.

ConditionScore
Flagged as compound het candidate by classification1.0
Heterozygous with another het variant in same gene0.8
Not heterozygous or no partner variant0.0

7. Age Relevance Score

Uses curated gene lists to prioritize age-appropriate disease genes. A BRCA1 variant scores high for an adult but low for a neonate. A CFTR variant scores high for a neonate but moderate for an adult. See Age-Aware Prioritization for the complete gene lists and scoring tables per age group.

Missing Data Handling

All component calculations handle missing data gracefully. When a predictor score is unavailable (NULL), that predictor is excluded from the weighted calculation rather than treated as zero. When gene constraint data is missing, the constraint score defaults to the gene's pLI value or 0.0. This ensures that variants with incomplete annotation are not unfairly penalized or promoted.