Quality control report for Example 50K

simulated SNP chip data

John Doe and Jane Doe

May 30, 2013

Abstract

This report encompasses the quality control summary for the Example50K simulated SNP chip data. A total of 83 samples were genotyped for54977 SNPs. Quality control was performed across samples, across snpsand on physical location. The results for each of these and the filteringcriteria used are discussed herein.

1 QC filtering results

Out of the 83 samples, 2 did not pass the filtering criteria (2.41%). From the54977 SNPs 4757 were excluded (8.65%). Out of the total 4563091 genotypes,502210 were excluded (11.01%). Filtering criteria consisted of QC metrics acrossSNPs, across arrays and on the physical mapping as detailed in the followingsections.

Table 1 summarizes the number of SNPs and samples rejected for each QCcriterion. Note that many of these overlap across criteria, thus the final numbersare not simply a sum of the rejection numbers for each criterion.

The correlation criterion for samples was not used to reject samples but sim-ply to flag potential replicates which should be checked before further analyses.Correlation includes SNPs and samples flagged as bad which makes samples lesssimilar than they should be. The correlation matrix should be used only for QCpurposes. For downstream analysis the GRM constructed after data filteringshould be used.

2 SNP statistics

In this section the descriptive statistics for the dataset on a per SNP basis arediscussed. Figures 1 and 2 illustrate the difference between good and bad qualitygenotypes.

2.1 SNP call rates

The number of SNPs with a call rate higher than 99.5% was 74.3% (Table 2and Figure 3). As a rule of thumb around 90% of the snps would be expectedto have a call rate above 99.5% and less than 2% would have call rates under90%. In some cases the bulk of the data may be just below, in the 0.99-0.995

1

Table 1: Summary of SNPs and samples rejected for each QC criterion.

SNP criteria number>5 percent genotyping fail 1432

median GC scores <0.5 2078all GC scores 0 650

GC <0.5 in less than 90 percent samples 2557100 percent homozygous 178

MAF <0.01 129heterozygosity 3SD 6

Hardy-Weinberg at 1e-15 131

sample criteria numbercall rates <0.9 2

correlation >0.98 0heterozygosity 3SD 0

mapping criteria numberChromosome 0 317Chromosome X 1502Chromosome Y 66

● ● ● ●● ● ●●● ●

●●

●●●●● ● ●●● ●●●●● ● ●●●

● ●●●●

0.0 0.1 0.2 0.3

0.0

0.1

0.2

0.3

0.4

snp32129

x

y

● (35) (43) (5)

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0 20 40 60 80

0.0

0.2

0.4

0.6

0.8

1.0

NA or 0:0 gc<0.5:0 gc>=0.5:83

samples

gc s

core

allelic frequencies

−: 0 A: 0.68 B: 0.32

0.0

0.1

0.2

0.3

0.4

0.5

0.6

MIS AA AB BB AA AB BB

genotypic frequenciesHW p−value: 0.1348

MIS=missing, left expected, right observed

010

2030

40

Figure 1: Example of a good quality SNP. Top left: clustering for each genotype(non calls are shown as black circles). Top right: GC scores. Bottom left:non-calls and allelic frequencies (actual counts are shown under the histogram).Bottom right: genotypic counts, on the left hand side the expected counts andon the right the observed counts; the last block shows number of non-calls.

2

0.2 0.4 0.6 0.8 1.0

0.4

0.6

0.8

1.0

snp473

x

y

● (83) ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

0 20 40 60 80

0.0

0.2

0.4

0.6

0.8

1.0

NA or 0:83 gc<0.5:83 gc>=0.5:0

samples

gc s

core

allelic frequencies

−: 1 A: NaN B: NaN

0.0

0.2

0.4

0.6

0.8

1.0

MIS AA AB BB AA AB BB

genotypic frequenciesHW p−value: NA

MIS=missing, left expected, right observed

020

4060

80

Figure 2: Example of a bad quality SNP. Top left: clustering for each genotype(non calls are shown as black circles - here all samples). Top right: GC scores.Bottom left: non-calls and allelic frequencies (actual counts are shown under thehistogram). Bottom right: genotypic counts, on the left hand side the expectedcounts and on the right the observed counts; the last block shows number ofnon-calls.

3

Table 2: Call rates for SNPs.

rate count frequency<0.9 2557 0.047

0.9-0.95 568 0.0100.95-0.99 10988 0.200

0.99-0.995 0 0.000>=0.995 40864 0.743

0 20 40 60 80 100

0.0

0.2

0.4

0.6

0.8

1.0

distribution of call rates

proportion of snps

call

rate

freq

uenc

y

0.90.950.990.995

Figure 3: distribution of call rates per SNP.

band (see breakdown of call rates in 2). Note that this will not hold well if thereis ascertainment bias problems with the SNPs (i.e. SNPs selected for the chipderived from one population and the samples come from a very different one).In this dataset 3125 SNPs failed genotyping in over 5% of the samples (thesewere removed from the dataset). Note that the number of SNPs failed dependson the GC cutoff threshold – all SNPs below 0.5 are deemed to have failed (seefurther details in GC scores section).

2.2 GC scores

GC scores were filtered for a threshold value of 0.5. All calls under this valuewere discarded (note that this is specific for each snp on an individual sample).The dataset contained 650 SNPs where all GC scores were 0. A further 2557SNPs had a GC score over 0.5 in less than 90% of the samples. 30076 SNPshad a GC score of at least 0.9 for at least 90% of the genotypes. The mean GC

4

0.0 0.2 0.4 0.6 0.8 1.0

02

46

810

mean density distribution of GC scores

N = 54977 Bandwidth = 0.007809

Den

sity

100% GC=0: 650>90% GC<0.5: 2557>90% GC>0.9: 30076

Figure 4: Histogram of GC scores.

scores for this data is 0.859 and the median is 0.865. The distribution of GCscores is shown in Figures 4 and 5.

2.3 Minor allele frequency

The minor allele frequency (MAF) was calculated for each SNP. 178 SNPs arehomozygous for the locus. A further 129 had a MAF below 0.01 and were dis-carded. The distribution of MAFs is shown in figure 6. The average heterozy-gosity for the SNPs is 0.39 and the standard deviation is 0.137. A total 6 SNPswere detected as outliers (3SD from the mean and removed). Heterozygosity(He) and gene diversity (Ho) distributions are shown in figure 7.

2.4 Hardy-Weinberg equilibrium

Hardy-Weinberg (HW) equilibrium was calculated for each individual SNP usingan exact chi-square test with continuity correction. HW equilibirum could notbe determined for 2210 SNPS because these were either homozygous or hadno calls assigned. 127 SNPs had a p-value of 0. A p-value cutoff of 1e-15shows 131 SNPs out of HW equilibrium (note that this also includes SNPsthat would not be expected to be in HW equilibrium such as those on sexchromosomes, mitochondria, etc). Figure 8 shows the distribution of p-valuesfor HW equilibrium.

5

0.0_0.1

0.1_0.5

0.5_0.6

0.6_0.9

0.9_1

distribution of GC scores

Missing: 72441

0.0_0.1: 72441 (1.59%)0.1_0.5: 126873 (2.78%)0.5_0.6: 37491 (0.82%)0.6_0.9: 1850674 (40.56%)0.9_1: 2475612 (54.25%)

Figure 5: Pie plot of GC scores.

●

●

●

●

●

●

●

● ●

●

0.0 0.1 0.2 0.3 0.4 0.5

2000

3000

4000

5000

6000

7000

NAs: 2032 MAF=0:178 MAF<0.01: 307minor allele frequency

num

ber

of s

nps

Figure 6: Minor allele frequency distribution for SNPs.

6

0.0 0.2 0.4 0.6 0.8 1.0

02

46

8

Heterozygosity (Ho) and gene diversity (He) density plotHo − mean: 0.39 sd: 0.137 He − mean: 0.387 sd: 0.119

mean: black line / 3SD: red line / number of outliers: 6density

freq

uenc

y

HoHe

Figure 7: Heterozygosity distribution for SNPs. Note: standard deviations arebiased.

0 20 40 60 80 100

0.0

0.2

0.4

0.6

0.8

1.0

P−values for Hardy−Weinberg equilibrium

NAs: 2210 pval=0: 127 pval<1e−15:131proportion of SNPs

p−va

lues

pval=0pval<1e−15

Figure 8: P-value distribution and thresholds for Hardy-Weinberg equilibrium.

7

Table 3: Call rates for samples.

statistic valuenum samples 83

min 0.825max 0.961

mean 0.956<0.97 83<0.9 2

Table 4: Sample pairs with high correlations.

sample1 sample2 correlation

3 Array and sample statistics

In this section the descriptive statistics for the dataset on a per chip/samplebasis are discussed.

3.1 Sample call rates

Out of the total 83 samples, 81 samples had a call rate at or above 0.9% and0 samples had a call rate at or above 97%. The mean call rate across sampleswas 95.63%. An overview is given in Table 3.

3.2 Sample correlations

The average correlation between samples is 0.795. The statistic is useful toidentify replicates in the dataset and samples that show very divergent genotypesdue to quality problems (Figure 9). The minimum is 0.386 and the maximum is0.908. 0 samples have a correlation above 0.98. Figure 10 shows the distributionof correlations between samples. The sample pairs with high correlations aregiven in Table 4. Note: correlation herein is a simple Pearson correlation ofthe entire dataset without correcting for allelic frequencies or removing missingcalls (use the GRM for downstream analyses). For this reason, even replicatesamples will not have a perfect correlation of one (e.g. a given snp is called inone sample and missing in the replicate). A missing value of nine is used whichteases genotypes with problems quite strongly apart.

3.3 Sample heterozygosity

The average heterozygosity for the samples is 0.39 and the standard deviation is0.013. A total 0 samples were detected as outliers (3SD from the mean). Sampleheterozygosity is shown in figure 11.

8

Figure 9: Heatmap of correlations between samples.

0.4 0.5 0.6 0.7 0.8 0.9

010

2030

4050

Correlation between samples

corr

elat

ion

min: 0.386max: 0.908mean: 0.795median: 0.821>0.9: 2<0.1: 0

Figure 10: Correlations between samples.

9

●●●●●

● ●●●●●

● ● ●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●

●●●●

0.36 0.38 0.40 0.42

020

4060

80

Sample heterozygositymean: 0.39 sd: 0.013

mean: black line 3 SD: red line number of outliers: 0heterozygosity

sam

ple

Figure 11: Heterozygosity for samples. Note: standard deviations are biased.

10

location

HW

chi

squa

re

0

20

40

60

80

0 1 10 11 12 13

14 15 16 17 18

0

20

40

60

80

190

20

40

60

80

2 20 21 22 23 24

25 26 3 4 5

0

20

40

60

80

60

20

40

60

80

7 8 9 X Y

Figure 12: Hardy-Weinberg plotted against physical location for each chromo-some (unmapped SNPs also included).

4 Physical mapping summary

A summary of the mapping information for the chip is given in table 5. Phys-ical mapping plots for Hardy-Weinberg, MAF, GC scores and heterozygositystatistics are respectively shown in Figures 12, 13, 14 and 15. 1885 SNPs areon excluded chromosomes and were removed. Many SNPs on e.g. the X chro-mosome are, as would be expected, out of HW equilibrium. The key point is toobserve if any of the other chromosomes show a clear pattern of disequilibriumin any particular region. The same applies to MAF, GC scores and heterozy-gosity chromosomal plots - an indication of problems is a pattern in any givenregion.

11

Table 5: Summary of mapping information per chromosome. Second column isthe number of SNPs per chromosome. Columns min, max and mean are respec-tively the minimum distance between adjacent SNPs, the maximum distance andthe average distance.

chrom num min max mean0 317 0 0 01 6016 1753 898900 498102 5553 1936 898500 473903 5071 5326 872500 478704 2723 5393 407800 467005 2385 65 842200 489706 2624 5342 2937000 491907 2280 5303 1053000 476808 2084 5385 452200 470009 2168 5298 752500 46530

10 1871 3113 3419000 5033011 1202 5300 429300 5567012 1738 5288 953600 4954013 1718 5478 903000 5175014 1193 37 873200 5803015 1716 5414 1266000 5238016 1596 5691 424000 4838017 1439 5284 562700 5455018 1434 5366 538700 5019019 1257 5353 416900 5159020 1166 5373 683200 4768021 912 5363 1970000 6073022 1107 5312 2215000 4975023 1140 5385 688100 5820024 750 5514 343600 5929025 1016 5514 589600 4733026 933 915 1692000 53460X 1502 5314 2299000 85230Y 66 0 0 0

12

location

min

or a

llele

freq

uenc

y

0.0

0.1

0.2

0.3

0.4

0.50 1 10 11 12 13

14 15 16 17 18

0.0

0.1

0.2

0.3

0.4

0.519

0.0

0.1

0.2

0.3

0.4

0.52 20 21 22 23 24

25 26 3 4 5

0.0

0.1

0.2

0.3

0.4

0.56

0.0

0.1

0.2

0.3

0.4

0.57 8 9 X Y

Figure 13: Minor allele frequencies plotted against physical location for eachchromosome (unmapped SNPs also included).

13

location

med

ian

GC

sco

re

0.0

0.2

0.4

0.6

0.8

1.00 1 10 11 12 13

14 15 16 17 18

0.0

0.2

0.4

0.6

0.8

1.019

0.0

0.2

0.4

0.6

0.8

1.02 20 21 22 23 24

25 26 3 4 5

0.0

0.2

0.4

0.6

0.8

1.06

0.0

0.2

0.4

0.6

0.8

1.07 8 9 X Y

Figure 14: GC scores plotted against physical location for each chromosome(unmapped SNPs also included).

14

location

hete

rozy

gosi

ty

0.0

0.2

0.4

0.6

0.8

1.00 1 10 11 12 13

14 15 16 17 18

0.0

0.2

0.4

0.6

0.8

1.019

0.0

0.2

0.4

0.6

0.8

1.02 20 21 22 23 24

25 26 3 4 5

0.0

0.2

0.4

0.6

0.8

1.06

0.0

0.2

0.4

0.6

0.8

1.07 8 9 X Y

Figure 15: Heterozygosity plotted against physical location for each chromosome(unmapped SNPs also included).

15