Fig. 3.

Comparison of DEploid and existing tools (COIL, pfmix, BEAGLE and SHAPEIT). (a) Estimates for the number of strains present in each mixed infection (artificially mixed in the lab) as given by COIL and DEploid. (b) Comparison of the inferred effective number of strains of each mixture as given by pfmix and DEploid. (c) Relationship between strain proportions and haplotype inference accuracy in the experimental validation for DEploid and BEAGLE/SHAPEIT (only mixtures of two strains). We used Reference Panel V to deconvolute all 27 samples with default settings. Each point represents a deconvolved haplotype with 17 530 sites. Point shape refers to strain and colour indicates the method applied. We use LOESS smoothing to show the trend of error versus strain proportion. Top panel shows switch error rate, the middle panel indicates genotyping error rate and the bottom panel indicates genotyping error rate through strain dropout. Note that zero switch error is represented as points below one. In summary, we find that DEploid results for the number of strains and relative proportions in a mixture are comparable to those achieved by existing methods, while inferred haplotypes are considerably better than from other methods