FLB Computation¶

The Full Likelihood Bayes (FLB) factor quantifies cosegregation evidence for a genetic variant in a pedigree. Values greater than 1 support pathogenicity; values less than 1 argue against it.

\[ \text{FLB} = \frac{\prod_i P(\text{observations}_i \mid \text{heterozygous})} {\prod_i P(\text{observations}_i \mid \text{non-carrier})} \]

Pipeline Overview¶

HerediCalc computes the FLB in six steps:

incidence_source  →  trait_mapper  →  hazard_model
                                           ↓
                                    penetrance_model
                                           ↓
                               liability_assigner (per member)
                                           ↓
                                    flb_calculator  →  FLB

Each step is handled by a plugin. The default configuration uses:

Step	Plugin
Incidence data	`ci5_ix` (or another CI5 edition)
Trait mapping	`ci5_ix_hbopc`
Hazard rates	`annual_rate`
Penetrance	`victor`
Liability assignment	`victor_standard`
FLB calculation	`segregatr`

The `segregatr` FLB Calculator¶

HerediCalc delegates the FLB calculation to the R package segregatr via subprocess. This avoids re-implementing the Elston-Stewart algorithm and ensures numerical compatibility with the reference implementation.

What `segregatr` Receives¶

Two temporary TSV files are written for each computation:

Pedigree TSV — one row per pedigree member:

Column	Description
`individual_id`	Integer member ID
`father_id`	Father's ID, or 0 for founders
`mother_id`	Mother's ID, or 0 for founders
`sex_code`	1 = male, 2 = female
`is_affected`	1 / 0
`is_proband`	1 / 0
`affection_known`	1 / 0
`genotype`	`"het"`, `"hom"`, `"nc"`, or `"NA"`
`liability_class`	Zero-based index into the penetrance table

Penetrance TSV — one row per liability class, three columns: penetrance_nc, penetrance_het, penetrance_hom.

Each row stores the probability that a member in that liability class is in their observed state (affected with a specific disease in a specific age band, or unaffected up to their last-known age). See VICTOR Model Tracks for how these values are computed.

Execution¶

Rscript --vanilla compute_flb.R <pedigree.tsv> <penetrance.tsv> <allele_freq>

The R script calls segregatr::FLB(...) and returns a single JSON line: {"flb": <value>}. HerediCalc parses the value and cleans up the temp files. On error, temp files are preserved for diagnostics.

Liability Classes¶

Every pedigree member is assigned a liability class index by the liability_assigner plugin. The index maps to a row in the penetrance table:

Affected members: the row for their canonical disease and age-at-diagnosis band
Unaffected members: the row for their sex and age-last-contact band
Unknown-sex members: the uninformative slot (all genotype penetrances equal)

The victor_standard assigner implements this matching. The zero-based index is passed to segregatr as the liability_class column.

Allele Frequency¶

The allele frequency q (CRHF — cumulative risk haplotype frequency) enters the Hardy-Weinberg prior used by segregatr to weight heterozygote vs. homozygote carriers. It is loaded from the crhf_model sub-plugin (built-in: lookup) using the genetic_entity name from the pipeline config.

Interpreting FLB Values¶

FLB	Evidence
< 0.1	Against pathogenicity
0.1–1	Weak evidence against
1–8	Weak to moderate evidence for
8–350	Strong evidence for pathogenicity
≥ 350	Very strong evidence (PP1_Strong per ACMG/InSiGHT)

Exact thresholds depend on the classification framework (ACMG, InSiGHT, ClinGen). The FLB from a single family is combined with other evidence lines in a multi-factorial likelihood model.