On this page we detail the quality control (QC) pipeline for for the BipEx dataset. Further plots, the underlying code and a document summarising the pipeline can be found on the BipEx github repository.

We first summarise the collection of samples, splitting across cohorts and subtypes. In addtition to Bipolar cases, we also a collection of Schizophrenia cases that will serve as positive controls for our PTV burden analyses.

Location	Bipolar Disorder	Schizoaffective	Schizophrenia	Other	Unknown	Controls	Total
Aberdeen, UK	0	0	564	0	1	331	896
Amsterdam, NED	1,212	21	1	58	17	1,611	2,920
Baltimore, USA	380	0	0	8	0	126	514
Boston, USA	3,449	52	0	0	0	3,498	6,999
Cambridge, UK	0	0	0	0	0	2,851	2,851
Cardiff, UK	2,442	68	2,990	18	0	1,106	6,624
Dublin, IRE	180	11	29	3	0	9	232
Edinburgh, UK	885	6	304	0	0	64	1,259
London, UK	1,909	157	1,595	0	0	1,203	4,864
Stockholm, SWE	5,160	1	0	0	0	5,541	10,702
Umea, SWE	472	0	0	0	0	459	931
Wurzburg, GER	397	0	0	0	14	414	825
Total	16,486	316	5,483	87	32	17,213	39,617

Within the collection of Bipolar cases, we have subtype information: Bipolar 1, and Bipolar 2. Further splitting the cases and labelling Bipolar cases for whom we do not have subtype information available, we obtain the following numbers of samples in each subcategory:

Location	Bipolar Disorder 1	Bipolar Disorder 2	Bipolar Disorder NOS	Schizoaffective	Bipolar Total (including schizoaffective)	Controls	Total
Aberdeen, UK	0	0	0	0	0	331	331
Amsterdam, NED	1,032	169	10	21	1,233	1,611	2,844
Baltimore, USA	358	9	5	0	380	126	506
Boston, USA	2,122	390	576	52	3,501	3,498	6,999
Cambridge, UK	0	0	0	0	0	2,851	2,851
Cardiff, UK	1,518	772	67	68	2,510	1,106	3,616
Dublin, IRE	180	0	0	11	191	9	200
Edinburgh, UK	368	114	2	6	891	64	955
London, UK	1,309	372	0	157	2,066	1,203	3,269
Stockholm, SWE	2,364	1,753	905	1	5,161	5,541	10,702
Umea, SWE	320	149	3	0	472	459	931
Wurzburg, GER	216	159	15	0	397	414	811
Total	9,787	3,887	1,583	316	16,802	17,213	34,015

Additionally, for a subset of the available phenotype data, we also have information regarding psychosis diagnosis and age of onset according to three definitions and two age groupings. Restricting to the subset of cohorts for which we have psychosis information, the breakdown is as follows:

Location	Bipolar Disorder with Psychosis	Bipolar Disorder without Psychosis	Total
Boston, USA	667	172	839
Cardiff, UK	1,168	777	1,945
London, UK	1,095	584	1,679
Stockholm, SWE	2,161	2,394	4,555
Wurzburg, GER	59	338	397
Total	5,150	4,265	9,415

The age of onset categories we have data for are: age at first impairment, age at first symptoms, and age at first diagnosis. Each of these are then binned into two different partitions ‘24’ and ‘40’, according to the following rules:

‘24’ partition

Age of onset < 12 \(\rightarrow\) 0

Age of onset < 12-24 \(\rightarrow\) 1

Age of onset > 24 \(\rightarrow\) 2

‘40’ partition

Age of onset < 18 \(\rightarrow\) 0

Age of onset < 18-40 \(\rightarrow\) 1

Age of onset > 40 \(\rightarrow\) 2

The breakdown of the age of onset categories is as follows:

Location	Age First Impairment ‘24’	Age First Impairment ‘40’	Age First Symptoms ‘24’	Age First Symptoms ‘40’	Age at Diagnosis ‘24’	Age at Diagnosis ‘40’
Boston, USA	0	0	0	0	0	514
Cardiff, UK	1,308	1,308	1,199	1,199	0	0
London, UK	1,808	1,808	0	0	0	0
Stockholm, SWE	561	561	2,114	0	1,927	1,927
Total	3,677	3,677	3,313	1,199	1,927	2,441

Within each of these age of onset category splits, the age breakdown is then:

Age at first impairment

Location	Age First Impairment <12	Age First Impairment 12-24	Age First Impairment >24	Total ‘24’	Age First Impairment <18	Age First Impairment 18-40	Age First Impairment >40	Total ‘40’
Cardiff, UK	80	824	404	1,308	469	782	57	1,308
London, UK	78	978	752	1,808	446	1,188	174	1,808
Stockholm, SWE	26	256	279	561	135	355	71	561
Total	184	2,058	1,435	3,677	1,050	2,325	302	3,677

Age at first symptoms

Location	Age First Symptoms <12	Age First Symptoms 12-24	Age First Symptoms >24	Total ‘24’	Age First Symptoms <18	Age First Symptoms 18-40	Age First Symptoms >40	Total ‘40’
Cardiff, UK	215	757	227	1,199	684	481	34	1,199
Stockholm, SWE	253	1,148	713	2,114	0	0	0	0
Total	468	1,905	940	3,313	684	481	34	1,199

Age at first diagnosis

Location	Age First Diagnosis <12	Age First Diagnosis 12-24	Age First Diagnosis >24	Total ‘24’	Age First Diagnosis <18	Age First Diagnosis 18-40	Age First Diagnosis >40	Total ‘40’
Boston, USA	0	0	0	0	224	214	76	514
Stockholm, SWE	60	891	976	1,927	408	1,247	272	1,927
Total	60	891	976	1,927	632	1,461	348	2,441

Finally, the split by PI was as follows:

PI	Bipolar Disorder 1	Bipolar Disorder 2	Bipolar Disorder NOS	Schizoaffective	Bipolar Total (including schizoaffective)	Schizophrenia	Other	Unknown	Controls	Total
Andreas Reif	216	159	15	0	397	0	0	14	414	825
Andrew McQuillin	1,309	372	0	157	2,066	1,595	0	0	1,203	4,864
Bob Yolken	117	9	5	0	139	0	8	0	126	273
Danielle Posthuma	0	0	0	0	0	0	0	1	948	949
David St Clair	0	0	0	0	0	564	0	1	331	896
Derek Morris	180	0	0	11	191	29	3	0	9	232
Douglas Blackwood	368	114	2	6	891	304	0	0	64	1,259
Fernando Goes	241	0	0	0	241	0	0	0	0	241
Jordan Smoller	2,122	390	576	52	3,501	0	0	0	3,498	6,999
Michael ODonovan	0	0	0	11	11	2,986	1	0	0	2,998
Michael Owen	0	0	0	0	0	0	0	0	1,106	1,106
Mikael Landen	2,364	1,753	905	1	5,161	0	0	0	761	5,922
Nancy Pedersen	0	0	0	0	0	0	0	0	4,780	4,780
Nick Craddock	1,518	772	67	57	2,499	4	17	0	0	2,520
Roel Ophoff	1,032	169	10	21	1,233	1	58	16	663	1,971
Rolf Adolfsson	320	149	3	0	472	0	0	0	459	931
Willem Ouwehand	0	0	0	0	0	0	0	0	2,851	2,851
Total	9,787	3,887	1,583	316	16,802	5,483	87	32	17,213	39,617

For our QC pipeline, we first read in the .vcf file, split multiallelics, and remove sites with more than 6 alleles. After splitting muliallelics in the .vcf file containing 29,911,479 variants and restricting to these sites, we have 37,344,246 variants.

Initial genotype filtering

Our first step (after conversion of the joint called .vcf file to a hail matrix table) is to remove genotypes based on the following collection of criteria:

If homozygous reference, remove if at least one of the following is true:
- Genotype quality \(<\) 20
- Depth \(<\) 10
If heterozygous, at least one of the following is true:
- (Reference allele depth + alternative allele depth) divided by total depth \(<\) 0.8
- Alternative allele depth divided by total depth \(<\) 0.2
- Reference phred-scaled genotype posterior \(<\) 20
- Depth \(<\) 10
If homozygous variant, at least one of the following is true:
- Alternative allele depth divided by total depth \(<\) 0.8
- Reference phred-scaled genotype posterior \(<\) 20
- Depth \(<\) 10

Initial variant filtering

Remove variants that either:

Fall in a low complexity region
Fail variant quality score recalibration (VQSR)
Fall outside padded target intervals (50bp padding)
Are invariant after the initial GT filter

Filter	Variants	%
Variants with < 7 alleles	37,344,246	100.0
Failing VQSR	100,742	0.3
In LCRs	1,215,218	3.3
Outside padded target interval	27,119,165	72.6
Invariant sites after initial variant and genotype filters	3,117,961	8.3
Variants after initial filtering	6,829,373	18.3

Initial sample quality control

Filter	Samples	Bipolar cases	Controls	%
Initial samples in vcf	39,618	16,486	17,212	100.0
Unable to obtain both phenotype and sequence information	2	NA	NA	0.0
Unknown phenotype	32	NA	NA	0.1
Low coverage or high contamination	133	72	54	0.3
Samples after initial filter	39,451	16,414	17,158	99.6

We run the sample_qc function in hail and remove samples according to the following:

Sample call rate \(<\) 0.93
FREEMIX contamination (%) \(>\) 0.02
Percentage chimeras (%) \(>\) 0.015
Mean depth \(<\) 30
Mean genotype quality \(<\) 55

Thresholds used were based on plotting the distributions of these metrics. A full collection of plots can be found in the repository. Here we show boxplots with overlaid scatterplots of the above metrics, split by sequencing batch, and coloured by location. The threshold for exclusion is shown as a dashed line.

Filter	Samples	Bipolar cases	Controls	%
Samples after initial filter	39,451	16,414	17,158	100.0
Sample call rate < 0.93	185	124	53	0.5
% FREEMIX contamination > 0.02	268	146	104	0.7
% chimeric reads > 0.015	152	49	100	0.4
Mean DP < 30	20	5	12	0.1
Mean GQ < 55	56	28	25	0.1
Samples after sample QC filters	38,894	16,138	16,906	98.6

Following this step, we export high quality variants (allele frequency between 0.01 to 0.99 with high call rate (> 0.98)) to plink format and prune to pseudo-independent SNPs using --indep 50 5 2. This pruned set of SNPs feeds into the next few stages of the QC pipeline.

Sex imputation

We impute the sexes of the individuals with this pruned set of variants on the X chromosome, and create list of samples with incorrect or unknown sex as defined by:

Sex is unknown in the phenotype files
F-statistic \(>\) 0.6 and the sex is female in the phenotype file
F-statistic \(<\) 0.6 and the sex is male in the phenotype file

Here we show the distribution of the F-statistic, with the 0.6 threshold defining our sex impututation shown as a dashed line.

Filter	Samples	Bipolar cases	Controls	%
Samples after sample QC filters	38,894	16,138	16,906	100.0
Samples with sex swap	238	147	52	0.6
Samples after sex swap removal	38,656	15,991	16,854	99.4

IBD

Using the identity_by_descent method in hail, we evaluate \(\hat{\pi}\) between pairs of samples, and filter based on a threshold of 0.2 shown as a dashed line on the plot below.

We then create a sample list of patients such that no pair has \(\hat{\pi} >\) 0.2.

Filter	Samples	Bipolar cases	Controls	%
Samples after sample QC filters	38,894	16,138	16,906	100.0
Related samples for removal	1,716	792	688	4.4
Samples after IBD removal	37,178	15,346	16,218	95.6

PCA

We next perform a number of principal component analysis (PCA) steps to ensure that we have matched cases and controls in our cleaned dataset.

Initial PCA

We first run PCA on samples after removing relateds and those that passed initial QC, using the pruned set of variants.

PCA including 1000 genomes

Next, we included the 1000 genomes samples (minus the small subset of related individuals within 1000 geneomes), and rerun PCA after including those individuals. Plots of the first six principal components are shown below. 1000 genomes samples are coloured in dark blue.

We restrict to the European subset of individuals to perform analysis. To do this, we train a random forest on the super populations labels of 1000 genomes and predict the super population that each of the BipEx samples. We denote strictly defined European subset as those with probability \(>\) 0.95 of being European according to the classifier. BipEx samples are coloured by their assignment or unsure if none of the classifier probabilities exceeded 0.95 in the following plots.

Samples not assigned to the European cluster were removed from downstream analysis.

In addition, using a much looser definition of European, we restrict to US samples from MGH and Johns Hopkins, and run PCA. This enabled us to identify Ashkenazi Jewish clusters, and create a list of outliers (AJ or otherwise) for downstream removal or independent analysis.

Run also ran a further collection of PCAs on:

Strictly defined Europeans and Ashkenazi Jewish individuals
- Use Ashkenazi Jewish cluster to train a random forest and determine if there are further Ashkenazi Jews in the remainder of the dataset.
Strictly defined Europeans

Filter	Samples	Bipolar cases	Controls	%
Samples after IBD removal	37,178	15,346	16,218	100.0
PCA based filters	2,880	1,120	1,422	7.7
Samples after population filters	34,298	14,226	14,796	92.3

However, upon restriction to the European cluster and after removal of AJs, we find that we have a dense case-control matched collection of samples and so decide not to analyse Swedes, Finns and Europeans (excluding Finns and Swedes) separately.

Final variant filtering

For our final variant filtering step, we first restrict to samples in the strictly defined European subset, filter to the unrelated list, and filter out samples with incorrectly defined sex or unknown sex, and run variant QC. We then evaluate a collection of variant metrics and remove variants that satisfy at least one of:

Invariant site in cleaned sample subset
Call rate \(<\) 0.97
Control call rate \(<\) 0.97
Case call rate \(<\) 0.97
\(|\)Case call rate - Control call rate\(| >\) 0.02
\(p\)-value for Hardy Weinberg Equilibrium \(<\) 10^-6

The following plots show the 0.97 threshold for call rate and 0.02 threshold for difference in call rate between cases and controls respectively.

Filter	Variants	%
Variants after initial filter	6,829,373	100.0
Invariant sites after sample filters	1,051,421	15.4
Overall variant call rate < 0.97	737,072	10.8
Overall variant case call rate < 0.97	716,709	10.5
Overall variant control call rate < 0.97	743,659	10.9
Difference between case and control variant call rate < 0.02	232,341	3.4
Variants failing HWE filter	1,083,479	15.9
Variants after filters	5,104,759	74.7

After these steps we plot the resulting changes in metrics across the samples in our data set. Each of the following plots splits the data by sequencing data and colours the points based on location. The first collection of subplots in each figure shows the variant metrics before sample removal, with the lower collection of subplots showing the resultant change after our QC steps.

Final sample filtering

In this step we remove sample outliers after the variant cleaning in the previous step. Samples are removed if at least one of the following lies more that three standard deviations away from the mean:

Ratio of transitions to transversions
Ratio of heterozygous to homozygous variant
Ratio of insertions to deletions
Number of singletons

Filter	Samples	Bipolar Cases	Controls	%
Samples after population filters	34,298	14,226	14,796	100.0
Within batch Ti/Tv ratio outside 3 standard deviations	100	50	42	0.3
Within batch Het/HomVar ratio outside 3 standard deviations	150	66	58	0.4
Within batch Insertion/Deletion ratio outside 3 standard deviations	93	31	48	0.3
Within location n singletons outside 3 standard deviations	443	151	236	1.3
Samples after final sample filters	33,527	13,933	14,422	97.8

As a final step, we export common (allele frequency between 0.01 and 0.99) variants to plink format, prune, and evaluate final principal components for downstream analysis. The first six principal components are displayed below and coloured by case status.

After all of this data cleaning, we save the resultant hail matrix tables for downstream analyses.

The resultant composition of the samples was as follows:

Location	Bipolar Disorder	Schizoaffective	Schizophrenia	Other	Unknown	Controls	Total
Aberdeen, UK	0	0	521	0	1	322	844
Amsterdam, NED	1,116	19	1	57	17	1,359	2,569
Baltimore, USA	267	0	0	4	0	41	312
Boston, USA	2,434	31	0	0	0	2,544	5,009
Cambridge, UK	0	0	0	0	0	2,656	2,656
Cardiff, UK	2,108	65	2,489	17	0	1,006	5,685
Dublin, IRE	150	11	27	2	0	7	197
Edinburgh, UK	711	6	271	0	0	58	1,046
London, UK	1,731	144	1,476	0	0	1,082	4,433
Stockholm, SWE	4,609	1	0	0	0	4,530	9,140
Umea, SWE	441	0	0	0	0	426	867
Wurzburg, GER	366	0	0	0	12	391	769
Total	13,933	277	4,785	80	30	14,422	33,527

The bipolar subtype information of the curated samples is:

Location	Bipolar Disorder 1	Bipolar Disorder 2	Bipolar Disorder NOS	Schizoaffective	Bipolar Total (including schizoaffective)	Controls	Total
Aberdeen, UK	0	0	0	0	0	322	322
Amsterdam, NED	951	155	9	19	1,135	1,359	2,494
Baltimore, USA	254	6	4	0	267	41	308
Boston, USA	1,503	279	404	31	2,465	2,544	5,009
Cambridge, UK	0	0	0	0	0	2,656	2,656
Cardiff, UK	1,301	681	62	65	2,173	1,006	3,179
Dublin, IRE	150	0	0	11	161	7	168
Edinburgh, UK	317	94	2	6	717	58	775
London, UK	1,169	350	0	144	1,875	1,082	2,957
Stockholm, SWE	2,095	1,595	791	1	4,610	4,530	9,140
Umea, SWE	297	141	3	0	441	426	867
Wurzburg, GER	201	145	13	0	366	391	757
Total	8,238	3,446	1,288	277	14,210	14,422	28,632

Taking a look at the breakdown of samples for which we have psychosis diagnosis information available:

Location	Bipolar Disorder with Psychosis	Bipolar Disorder without Psychosis	Total
Boston, USA	454	122	576
Cardiff, UK	993	693	1,686
London, UK	978	551	1,529
Stockholm, SWE	1,925	2,135	4,060
Wurzburg, GER	55	311	366
Total	4,405	3,812	8,217

Finally, we can look at the breakdown of samples in each of the age of onset categories: