Quality Metrics

Quality metrics for haplotype frequency data

These are mean to characterize data so that thresholds can be consistently applied such as: only considering data with a RES_TRS_COUNT > x or only SAM_SIZE > y.

Many of these depend on having the genotypes, or some aspect of the genotypes, available.

QUAL_TYPE	GENOTYPE BASED	VALUE	DESCRIPTION
DIV_ALPHA		real, < 0	Exponent of Power Law fit to HTF distribution (This is called alpha in Slater et al. Power Laws for Heavy- Tailed Distributions)
DIV_50	X	integer	Number of haplotypes needed (in descending order of frequency) to have the cumulative sum be > 0.5 (Sample size sensitive!)
DIV_50_REL		Real, 0 <= x <= 1	Number of haplotypes needed (in descending order of frequency) to have the cumulative sum be > 0.5 divided by the number of HT
SAM_SIZE	X	integer	Number of GT
DIV_PGD	X	Real, 0 <= x <= 1	Population genetics diversity (1-sum f_i ^2 N/(N- 1)) where N = SAM_SIZE and f_i is the frequency of a specific haplotype
DIV_HEAVY_TAIL		Real, 0 <= x <= 1	a is an independence parameter of the Bayesian SHF model that describes how allele frequency products correlate with haplotype frequencies (also correlates with the fraction of nonzero categories) – From Yoram SHF MS
RES_TRS_COUNT		Real, 0 <= x <= 1	Average number of possible genotypes per individual
RES_TRS		Real, 0 <= x <= 1	Typing Resolution Score – Average sum of square of genotype probabilities (imputation method dependent; ideally imputed from the HF estimated)
RES_SHARE_AMBIG	X	Real, 0 <= x <= 1	Fraction of GT with a lower resolution than defined in the resolution tag
RES_MISS_LOCI	X	Real, 0 <= x <= 1	Fraction of GT with missing loci (separate qual_type per locus?)
DEV_HWE	X	Real	Deviation from HWE (using HWE with ambiguity method)
ERR_STD		Real, 0 <= x <= 1	Weighted average of standard errors across all haplotypes
ERR_SAMP_80_100		Real	Laurent, Excoffier “If” between frequencies derived from 100% set and 80% training set
SUM_FREQ_GAP		Real	Sum of haplotype frequencies for unobserved haplotypes that are expected in population by SHF model
ERR_OFFSET		Real, 0 <= x <= 1	1-sum f_i (Difference between predicted full HF distribution using SHF versus actual including test set?)
LD_MEASURE		Real	Define in Method section – (Where is LD measured for quality?)
KFOLD_IMPUTE	X	Real, 0 <= x <= 1	% of imputable GT in 20% test set from HT generated in 80% training set
KFOLD_PRED_ACTUAL	X	Real, 0 <= x <= 1	Divergence between predicted and actual with Log Loss function (for test set predictions on simulated lower-resolution typings)
KFOLD_N		integer	Number of independent training-test folds (k)