Pedigree reconstruction from poor quality genotype data

Marker genotype data could suffer from a high rate of errors such as false alleles and allelic dropouts (null alleles) in situations such as SNPs from low-coverage next-generation sequencing and microsatellites from noninvasive samples. Use of such data without accounting for mistyping properly could lead to inaccurate or incorrect inferences of family relationships such as parentage and sibship. This study shows that markers with a high error rate are still informative. Simply discarding them could cause a substantial loss of precious information, and is impractical in situations where virtually all markers (e.g. SNPs from low-coverage next-generation sequencing, microsatellites from noninvasive samples) suffer from a similarly high error rate. This study also shows that some previous error models are valid for markers of low error rates, but fail for markers of high error rates. It proposes an improved error model and demonstrates, using simulated and empirical data of a high error rate (say, >0.5), that it leads to more accurate sibship and parentage inferences than previous models. It suggests that, in reality, markers of high error rates should be used rather than discarded in pedigree reconstruction, so long as the error rates can be estimated and used properly in the analyses.