RFMix

RFMix is a software tool for local ancestry inference in admixed populations. It uses a discriminative modeling approach based on random forests to assign the ancestral origin of chromosomal segments along the genome, using phased haplotypes from reference populations. The method was introduced to provide rapid and robust local ancestry calls in individuals with mixed ancestry and has become widely used in population genetics and disease studies.

The methodology combines a classifier and a genome-wide smoothing step. A random forest classifier is trained on labeled reference haplotypes from multiple ancestral populations, using a window of nearby markers to predict the ancestry of target haplotype segments. The trained model then assigns ancestry probabilities to each position or window in the target individuals. To maintain consistency across the genome and account for recombination, an additional step using a hidden Markov model or related smoothing framework integrates information across adjacent positions and incorporates a recombination map.

Inputs to RFMix include phased genotype data for the target individuals, phased reference panels representing the ancestral populations, and a genetic map. Optional inputs may include population labels for the reference samples and sample weights. Outputs consist of local ancestry calls at each genomic position or window for every individual, along with posterior probability estimates for each ancestry. These results enable downstream analyses such as admixture mapping, studies of fine-scale population structure, and control for local ancestry in association studies.

Limitations include dependence on accurate and representative reference panels, potential biases when ancestral populations are underrepresented or absent, and the requirement for phased data. Relatively high computational demands exist, though RFMix is designed to be faster than many alternative local ancestry methods.