How the breeding tools calculate

• Each locus is inherited on its own, unless loci share a linkage group. Linked loci on an autosome travel together as one block.
• Linked loci are read in coupling phase: an animal's mutant alleles sit on one chromosome. An animal typed as a carrier at one linked locus but not another is used as recorded, and the result says so.
• X-linked loci: sons take their mother's X and their father's Y; daughters take one X from each parent. Y-linked loci pass from father to son only.
• At a locus whose assay shows only present or absent, most often a transgene, a carrier counts as one copy (hemizygous) unless you choose two.
• Probabilities are exact fractions with a power of two below the line, such as 1/8 or 3/32, shown with their percent. The offspring of one cross add up to exactly 1.
• A locus where either parent has no genotype is left out of the result and named. The calculator never guesses a missing genotype.

The inbreeding coefficient F of a pair's offspring is Wright's coefficient, from three generations of recorded parents: the sum, over every ancestor the sire and dam share, of (1/2)^{n + m + 1}, where n and m are the generations from the sire and from the dam to that ancestor. A parent paired with its own offspring counts as 0.25.

Full siblings give 0.25, half siblings 0.125 and first cousins 0.0625. By default the planner leaves out pairs at 0.25 or more; a lab can choose 0.125, 0.0625 or no limit. Siblings bred within a plan count as 0.25. A pair with no recorded parents counts as unrelated in the app; the free public planner has no pedigree, so its pairs' F is unknown.

• Goal pups per litter: mu = share x litter size x (1 - pre-weaning loss), where the share is the fraction of pups with the goal's genotype and sex.
• Litters needed: the smallest whole number L with P(Poisson(L x mu) >= count) >= 0.8, so a plan reaches its count with a probability of at least 80%.
• Litters per pair: 1 + floor((window - pairing to first litter) / interval between litters), where the window runs from the pair's start to the last birth that can still be weaned and genotyped by the date.
• Pairs: ceil(L / litters per pair / share of pairs that produce litters). Pairs that start on different days each count their own litters.
• Completion: the genotyping date of the litter after which the expected goal pups reach the count with a probability of 0.8.
• Plans run up to three generations: animals on hand, then the pups of one step bred in the next. A bred animal joins a pair once it is expected (with probability 0.8) and old enough to breed.
• Breeders are taken in order of lowest F, then the youngest female, the youngest male and the animal ID; each animal once per plan. Up to the 20 youngest animals of each genotype are used.
• Cages: each pair holds a cage from its start until the plan completes; each litter's weaned pups wait for genotyping in 2 x ceil(pups / 2 / adults per cage) cages, one set per sex; the goal animals stay, per sex, until the date. Cage-days add up the cages in use each day, and cost is cage-days times the lab's cost per cage per day.
• When a goal names a protocol, a plan whose pups would take the protocol past its approved number is left out.
• Plans are ranked on five components, each scaled from 0 to 1 across the plans found: goal pups per litter, days to completion, pairs, cage-days and the most related pair. You set their weights; ties go to the earlier completion, then fewer pairs.

Every default the planner uses, with where it comes from. Your colony's own litters replace a default per strain once there are enough: litter size (pups born alive) from 5 recorded litters, the interval between litters from 5 intervals of the same pairs, and pre-weaning loss from 5 weaned litters, all born in the last 24 months. Dead-at-birth pups come off litter size once and are not counted again as losses. The planner names the source of every number it shows.

1. [1] JAX colony planning. The Jackson Laboratory, "Colony Planning" (read 2026-09-26). https://www.jax.org/jax-mice-and-services/customer-support/technical-support/breeding-and-husbandry-support/colony-planning
2. [2] Silver 1995. Silver LM. Mouse Genetics: Concepts and Applications. Oxford University Press, 1995. Table 4.1, "Reproductive characteristics of some important inbred strains". https://www.informatics.jax.org/silver/tables/table4-1.shtml
3. [3] McGill 2020. McGill University Comparative Medicine and Animal Resources Centre, "Module 5: Introduction to Rodent Breeding Colony Management" (Laboratory Animal Biomethodology Workshop handout, October 2020). https://www.mcgill.ca/research/files/research/breeding_module_handout_oct_2020.pdf
4. [4] Morello et al. 2020. Morello GM, Hultgren J, Capas-Peneda S, et al. "High laboratory mouse pre-weaning mortality associated with litter overlap, advanced dam age, small and large litters." PLoS One 2020;15(8):e0236290. https://doi.org/10.1371/journal.pone.0236290

Each plan run returns the engine version and an inputs hash: the SHA-256 of the canonical JSON of the engine version, the goal, the options, every parameter, each candidate animal with its genotype and the days it can breed, the F values, protocol use, cage capacity and the date. The same inputs always give the same plans and the same hash, so a plan can be checked later against the records it was made from.

In the cross planner you can describe the animals you need in your own words. A language model turns the words into a draft of the planner's form, using the loci your lab has recorded; anything it cannot match to a locus is listed for you, never invented. You check the form and press Plan. The model never computes a probability, count, date or cost. When AI is off, the box is hidden and the form works the same.

The anomaly detector and the colony assistant are in development; this page will describe their methods when they ship.