lshmm

lshmm is a Python library for prototyping, experimenting, and testing implementations of algorithms using the Li & Stephens (2003) Hidden Markov Model.

Usage

Inputs

Data

• Sample and/or ancestral haplotypes comprising a reference panel.
• Query haplotypes.

In the haploid mode, the alleles in haplotypes can be represented by any integer value (besides -1 and -2, which are special values). In the diploid mode, the genotypes (encoded as allele dosages) can be 0 (homozygous for the reference allele), 1 (heterozygous for the alternative allele), or 2 (homozygous for the alternative allele). Currently, multiallelic sites are supported in the haploid mode, but not the diploid mode.

Note that there are two special values NONCOPY and MISSING. NONCOPY (or -2) represent non-copiable states, and can only be found in partial ancestral haplotypes in the reference panel. MISSING (or -1) representing missing data, and can be found only in query haplotypes.

Parameters

• Recombination probabilities.
• Mutation probabilities.

Models and algorithms

• Haploid LS HMM
  • Forward-backward algorithm
  • Viterbi algorithm
• Diploid LS HMM
  • Forward-backward algorithm
  • Viterbi algorithm

Features

• Scaling of mutation rate by the number of distinct alleles per site.
• Non-copiable state in the reference panel (NONCOPY).
• Missing state in the query (MISSING).
• Multiallelic sites (haploid only).