| BEAGLE v.4.1 (Browning and Browning 2016)
|
Li and Stephens (2003) haplotype frequency model with parsimonious state-space |
HMM calculations restricted to clusters of markers (aggregates) within small regions |
| BEAGLE v.4.0 (Browning and Browning 2007, 2009) |
Iterates between building a suffix tree from current-guess haplotypes and updating haplotypes based on probabilities from the tree. The suffix tree resembles a “cluster-haplotype model,” but has a variable number of clusters depending on LD in each region |
Suffix tree is pruned to produce parsimonious representation of the data |
| fastPHASE (Scheet and Stephens 2006) |
EM algorithm that iterates between fitting hyper-parameters of a “haplotype-cluster HMM” and running forward-backward algorithm in fitted HMM to get imputed genotypes |
Hidden state modeled by haplotype clusters (20 clusters good empirically) |
| IMPUTE (v.1) (Marchini et al. 2007) |
HMM similar to Li-Stephens model. Forward-backward algorithm to compute hidden state probabilities, then analytically integrate over all hidden states |
Can restrict computation to reference panel haplotypes |
| IMPUTE2 (v.2) (Howie et al. 2009, 2011) |
MCMC iterating between phasing and imputing. Phasing done with IMPUTE v.1 HMM (HMM forward path sampling). Imputation done by haploid HMM (HMM forward-backward) |
Only subset of haplotypes with smallest Hamming distance to current-guess haplotypes in phasing step |
| MaCH (Li et al. 2010) |
HMM similar to Li-Stephens model. Iteratively update phase of each individual based on haplotypes of other individuals (HMM forward-backward). Additionally update HMM hyper-parameters at every iteration |
Only a subset of haplotypes randomly selected |
| MiniMac (Howie et al. 2012) |
Fast implementation of MaCH model using prephased data |
NA (phasing precomputed) |
| MiniMac2 (Fuchsberger et al. 2015) |
Computational speed-ups to the MiniMac software |
NA |
| SNPTools (Wang et al. 2013) |
“Constrained” Li-Stephens method |
Only four parental haplotypes, selected based on metropolis-hastings MCMC sampling |
| GeneImp |
Haplotype-pair sequence within window is exact copy of reference haplotype-pair. Reference haplotype-pairs compatible with genotype likelihoods are analytically integrated over |
Only a subset of “filtered” reference haplotypes |
