> Home

Introduction

Misconducts in the STAP cell research

A new type of induced pluripotent cells, STAP cells, was reported in two Nature papers by Dr. Obokata in RIKEN CDB and the others in 2014. Immediately after the publication, misuses of the data in the papers were pointed out by several readers. RIKEN set up Fraud Investigation Committees for the STAP papers. After the interviews and investigations on the authors of the STAP papers, the comittees concluded that there was misconduct by Dr. Obokata including total 3 fabrications and 1 falsification of data [ref1]. For the other figures in the papers, Dr. Obokata did not show any original data even after repeated requests from the committee although most of the figures were created by her. The committee reported that "There are few records of experiments. There are some experiments that we cannot tell they really were conducted (althogh we cannot tell they were not conducted either). "

Concurrently with this investigation on the research process, the committe reviewed the results of the scientific investigation of the samples obtained in the STAP study. The scientific investigation team in RIKEN analized the genotypes of STAP stem cells [ref4]. They found that some of them were different from those of mice used in the study. An obtained genotype suggested that some STAP stem cells were derived from ES cells that had been established for other reseach purposes. They finally did whole genome seuquencings of the selected STAP related cells and mice that were related to the STAP study. Total 15 genomes were sequenced and the NGS data were pulished on public database [ref2].

The committee concluded from the analysis that STAP cells used in a teratoma expeiment, STAP cells used for ChIP-seq, STAP stem cells, and FI stem cells were all derived from pre-existing ES cells established in Wakayama lab in RIKEN CDB, where Dr. Obokata and Dr. Wakayama mainly did most of their STAP cell experiments. The detail of the investigation was published as Nature B.C.A. by the original scientific investigation team in 2015 [ref3]. Their conclusion that STAP cells are derived from ES cells is based on similarities found in the genome structure, which cannot be explained if they are from different sources.

My goal (ongoing)

My specific goal of this analysis of published NGS data is to confirm their results and their conclusion that STAP cells are derived from ES cells. My interest is whether the difference between the ES cells (e.g., FES1) and STAP cells are explained by mutation accumulation that occured during cell culturing. Although this explanation has been suggested in the Nature B.C.A. paper, it lacks the quantitative evaluation and modeling of this process.

Genome sequence (DRA002862 in DDBJ)

Instruments: Illumina HiSeq 2500

Library construction: TruSeq DNA PCR-free;

Note: the sizes of the NGS reads are very big; .fastq - 67G byte, .fastq.gz - 20G byte, .fastq.bz2 - 15G byte

Note: There are 2 runs of NGS data for each cell. I don't know whether they are biological duplicates or technical duplicates.

Note: I can not share .bam files here because their sizes are too large (30-40GB) to store in servers I have rented.

Run table

FLS3 (ftp)

established at 1/31/2012-2/2/2012

STAP stem cell

Run nameforward readreverse readVCF
DRR028632 1 (FastQC) 2 (FastQC) 320MB
DRR02863312

CTS1 (ftp)

established at 5/25/2012 or 7/9/2012

FI stem cell

Run nameforward readreverse readVCF
DRR028642 1 (FastQC) 2 (FastQC) 320MB
DRR02864312

129/GFP ES (ftp)

unknown; found in Obokata's box in a freezer in Obokata lab.

Unknown type, possiblly ES cell

Run nameforward readreverse readVCF
DRR028638 1 (FastQC) 2 (FastQC) 320MB
DRR028639 1 2

FES1 (ftp)

129B6GFP1 FES(male); frozen at 12/7/2005

ES cell

Transgene: Acr3-EGFP/CAG-EGFP derived from transgenic mouse created in Okabe lab. (strain B6).

estabilished from F1 (129X1 x B6N) mouse

Run nameforward readreverse read BAMVCFIGV
DRR028646 1 (FastQC) 2 (FastQC) 320MB View
DRR028647 1 (FastQC) 2 (FastQC) stats. 320MB

FES2 (ftp)

129B6GFP2 FES(male); frozen at 12/7/2005

Run nameforward readreverse readVCF IGV
DRR028652 1 (FastQC) 2 (FastQC) 300MB View
DRR028653 1 2

ntESG1 (ftp)

129B6F1G1; frozen at 8/3/2007; named ntES1 in DDBJ

nuclear transfer ES cell

Transgene: Acr3-EGFP/CAG-EGFP derived from transgenic mouse created in Okabe lab. (strain B6).

estabilished from F1 (B6N x 129+Ter) mouse

Run nameforward readreverse read BAMVCF
DRR028644 1 (FastQC) 2 (FastQC) stats. 310MB
DRR028645 1 2

ntESG2 (ftp)

129B6F1G2; frozen at 1/20/2005; named ntES2 in DDBJ

Run nameforward readreverse readBAMVCF
DRR028650 1 (FastQC) 2 (FastQC) stats. 300MB
DRR028651 1 2

Mouse reference genome (GRCm38)

Ensembl

strain C57BL/6J

file size: .fa = 4.4G byte, .fa.gz = 0.8G byte

$ wget ftp://ftp.ensembl.org/pub/release-74/fasta/mus_musculus/dna/Mus_musculus.GRCm38.74.dna.toplevel.fa.gz

Transgenes

see my analysis results

CAG-GFP

CAG-EGFP, CAGGS-EGFP, CX-EGFP

EcoRI-EcoRI fragment of PCR product (primers: 5'-ttgaattcgccaccatggtgagc-3' and 5'-ttgaattcttacttgtacagctcgtcc-3') using pEGFP-C1 as a template was introduced into pCAGGS vector (CAG + pA signal) (ref)

3.2-kb SalI/BamHI fragment from pCAGGS-EGFP was used when creating Tg mouse

EGFP sequence from pEGFP-C1 (Clontech)

CAG promoter (1.7kb) + EGFP (0.7kb) + rabbit beta-globin pA and vector (0.7kb) = 3.1kb

Arc-GFP

Acr3-EGFP (ref1)

mouse acrosin promoter and N-terminal coding region (2.4kb) + EGFP (0.7kb) + rabbit globin pA = 3.7kb (ref2)

XbaI-HindIII fragment was used when creating Tg mouse

The original paper (ref1) described that Acr3-EGFP transgene construct used bovine growth hormone polyA signal (bGH pA). But ref2 uses rabbit globin polyA signal perhaps from pCAGGS vector. My personal investigation indicated that FES1 genome has no bGH pA sequence but has rabbit polyA signal sequence.

Oct4-GFP

GOF18-EGFP

GOF mouse ;18-kb genomic fragment of mouse Oct4 gene containing Oct4 pomoter region **

Mouse SNPs Data

variations relative to the C57BL/6J mouse reference genome

Mouse Genomes Project (.bam files)(.vcf files)

from Methods from the analysis paper

For comparison to 129/Sv SNPs, we used commonly called SNPs in 129P2/OlaHsd (EMBL ENA: ERP000034), 129S1/SvImJ (EMBL ENA: ERS076385), and 129S5SvEvBrd (EMBL ENA: ERP000036) but not called in C57BL/6NJ (EMBL ENA: ERS076384) provided by the Wellcome Trust Sanger Institute (http://www.sanger.ac.uk/resources/mouse/genomes/).

129P2/OlaHsd (.vcf.gz, 240M byte; .vcf, 1.9G; contains 6x10^6 SNPs)
$ wget ftp://ftp-mouse.sanger.ac.uk/current_snps/strain_specific_vcfs/129P2_OlaHsd.mgp.v5.snps.dbSNP142.vcf.gz
$ gzip -d ...vcf.gz
$ bgzip ...vcf
$ bcftools index ...vcf.gz

Update: I used a compiled version of VCF file obtained from ftp://ftp-mouse.sanger.ac.uk/REL-1303-SNPs_Indels-GRCm38/mgp.v3.snps.rsIDdbSNPv137.vcf.gz in later analyses. This file contains the alleles of all the inbed strains sequenced in Mouse Genomes Project.

Tools

see related file types

FastQC

A quality control application for high throughput sequence data

Babraham Bioinformatics; v0.11.5

written in Java

$ wget ...
$ unzip ...
$ sudo apt-get install default-jre

BWA

alignment via Burrows-Wheeler transformation (ref)

Heng Li; version 0.7.12-r1039

outputs .sam file

$ sudo apt-get install bwa

SAMtools/BCFtools

Tools for alignments in the SAM format

version 0.1.19-

for variation detections (SNPs, short INDELs, ...)

$ sudo apt-get install samtools

Later on, I used the latest version 1.3.1 for my analysis.

$ sudo apt-get update
$ sudo apt-get install gcc
$ sudo apt-get install libncurses5-dev
$ sudo apt-get install zlib1g-dev
$ wget https://github.com/samtools/samtools/releases/download/1.3.1/samtools-1.3.1.tar.bz2
$ tar jxvf samtools-1.3.1.tar.bz2
$ cd samtools-1.3.1 
$ ./configure
$ sudo apt-get install make
$ make
$ sudo make install
$ wget https://github.com/samtools/bcftools/releases/download/1.3.1/bcftools-1.3.1.tar.bz2
$ ...
$ sudo apt-get install gcc
$ sudo apt-get install zlib1g-dev
$ make
$ sudo make install

Logouts of the SSH session were required to update /usr/local/bin/samtools and bcftools after their installations.

tabix/bgzip was also useful.

$ sudo apt-get install tabix

Integrative Genomics Viewer (IGV)

written in Java and works in GUI environment

I used this software in Windows desktop environment.

Data analysis

automation

alignment (140h)

$ ~/FastQC/fastqc -o reports -t 1 <*.fastq.gz>
took about 0.5h
$ bwa index -p GRCm38_index Mus_musculus.GRCm38.74.dna.toplevel.fa.gz
took about 1.5-2h
$ bwa aln -n0.02 ../../genome/GRCm38_index DRR028646_1.fastq > DRR028646_1.sai
$ bwa aln -n0.02 ../../genome/GRCm38_index DRR028646_2.fastq.gz > DRR028646_2.sai
took about 50-60h each; consumed max. 5G byte; bwa aln for .fastq.bz2 didn't work in my environment for unknown reasons; .sai = 5-6G byte
$ nohup bwa sampe -a2000000 ../../genome/GRCm38_index DRR028646_1.sai DRR028646_2.sai DRR028646_1.fastq.gz DRR028646_2.fastq.gz > DRR028646.sam 2> stderr.log &
took about 8h; consumed max. 7G byte; .sam = 130G byte
$ nohup bwa sampe -a2000000 ../../genome/GRCm38_index DRR028646_1.sai DRR028646_2.sai DRR028646_1.fastq.gz DRR028646_2.fastq.gz 2> stderr.log \
	| samtools view -bS - > DRR028646.bam 2> stderr.2.log &
took about 8.5h; consumed max. 7G byte; .bam = 44G byte
$ nohup samtools sort DRR028646.bam -o DRR028646.sort.bam &
took about 2h; consumed around 1G byte; sorted .bam = 32G byte
Multithread option (e.g., "-@6") worked well in this step.
$ mv DRR028646.sort.bam DRR028646.bam
$ samtools index DRR028646.bam
took about 10 min

variant calling (20h)

$ nohup samtools mpileup -uf ../../genome/Mus_musculus.GRCm38.74.dna.toplevel.fa.gz DRR028646.bam 2> stderr.log \
	| bcftools call -vc -Oz > DRR028646.vcf.gz 2> stderr.2.log &
took about 18h; .vcf.gz size = 320M byte

Note: the idexing of gzipped .fa.gz is not supported (only bgzip compression is supported; bgzip is included in tabix package).

Note: the usage of bcftools is very different from that of older versions. Most of the examples you can find in the web are obsolete.

$ bcftools stats DRR028646.vcf.gz > vcfstats
accessionnamenumber of SNPsnumber of INDELs
DRR028532FLS3 5200551804857
DRR028542CTS1 5210480814377
DRR028638129/GFP ES5241069828393
DRR028644ntESG1 5018359789025
DRR028646FES1 5208496803758
DRR028650ntESG2 4942827761603
DRR028652FES2 5040101761845
relative to GRCm38 (total bases = 4.34x10^9 bp; total chromosome bases = 2.63x10^9 bp)

SNP analysis (1st. trial - obsolete)

Methods


Multidimensional scaling (MDS)

Pairwise distances of whole genome consensus sequences (based on Kimura-1980, Methods in prep.). Smaller distance means higher similarity.
I used all the SNVs detected in the automated process, which potentially contain a lot of false positives.
Note: I found that the distance that I calculated from all the SNVs detected in FES1 replicates (DRR028646 and DRR028647) was 3.03 x 10^-4. Distances around this value threfore should be considered to indicate "background" differences among genetically identical cell lines in this analysis. For more precise comparison of similarities between the cells, it will be required to filter out a lot of false positive SNVs by using some criteria.

the idea of MDS

SNP analysis (2nd. trial)

I this 2nd. trial, I use techniques learned from my experiences on reproducing the figures of Endo-2014 to obtain more reliable results.

Extraction of SNPs that can distinguish strain 129 from B6

extract-129B6-SNPs.py by Kanso

import sys

def main():
	# ouput in VCF format
	
	col_chr = 0
	col_pos = 1
	col_SNPid = 2
	col_ref = 3
	col_alt = 4
	col_qual = 5
	col_filter = 6
	col_info = 7
	col_129P2 = 10-1
	col_129S1 = 11-1
	col_129S5 = 12-1
	col_C57B6NJ = 17-1
	col_strains = [col_C57B6NJ, col_129P2, col_129S1, col_129S5]
	
	lineno = 0
	
	with sys.stdin as fin: #use as a filter
		for line in fin:
			li = line.rstrip()
			if (len(li) >= 2) and (li[0] == '#') and (li[1] == '#'):
				continue
			ls = li.split("\t")
			lineno = lineno + 1
			if lineno == 1: #check file header
				labels = { col_chr:"#CHROM", col_pos:"POS", col_SNPid:"ID", col_ref:"REF", col_alt:"ALT", col_qual:"QUAL", col_filter:"FILTER", col_info:"INFO", col_129P2:"129P2", col_129S1:"129S1", col_C57B6NJ:"C57BL6NJ" }
				for col in labels.keys():
					if ls[col] != labels[col]:
						print >> sys.stderr, "column '" + labels[col] + "' could not be found in the input VCF file"
						return 2
				print "#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO"
			else:
				genotypes = [(ls[i].split(":"))[0] for i in col_strains]
				if (genotypes[0]==".") or (genotypes[1]==".") or (genotypes[2]==".") or (genotypes[3]=="."):
					continue
				if genotypes[1]==genotypes[2]==genotypes[3]:
					g0 = genotypes[0].split("/")
					g1 = genotypes[1].split("/")
					
					if (g0[0]==g0[1]) and (g1[0]==g1[1]) and (g0[0]!=g1[0]):
						print "%s\t%s\t%s\t%s\t%s\t.\t.\t%s" %(ls[0],ls[1],ls[2],ls[3],ls[4],g0[0])
	return 0

if __name__ == '__main__':
	main()
$ cd stap/snps/dbSNP
$ zcat mgp.v3.snps.rsIDdbSNPv137.vcf.gz | python ~/extract-129B6-SNPs.py > 129B6-SNPs.vcf
It extracted the information of 5030545 SNPs from a SNPs database.
129B6-SNPs.vcf.gz (44MB)

Counting alles of the extracted SNPs in .bam files

I used snpexp. For the usage and installation of this software, see my report on trisomy-8 detection.

$ cd stap/reads/FES1
$ ~/snpexp/src/snpexp -V ~/stap/snps/dbSNP/129B6-SNPs.vcf DRR028646.p.MD.bam > DRR028646.p.MD.129B6.snpexp.out 2>/dev/null
DRR028646.p.MD.129B6.snpexp.out.gz (56MB)

Update: I found the original version of snpexp has a bug in the routine for allele counting. Therefore I use a fixed version snpexp2 here. I also use the original (no extractions of properly mapped reads) .bam files in this section for simplification. I have checked the effects of the extractions only sligtly change the resultant .bam files and there are no apparent differences in the visualized alllele distributions in the next section.

$ cd stap/reads/FES1
$ ~/snpexp/src/snpexp2 -V ~/stap/snps/dbSNP/129B6-SNPs.vcf DRR028646.bam > DRR028646.129B6.snpexp2.out
DRR028646.129B6.snpexp2.out.gz (52MB)

Visualization of the positional distribution of allele frequencies

visualization parameters:

full resolution

Detailed results

Allele frequency distribution for total chromosome


Dashed lines represent thresolds for homo/hetero discrimination.

filtering SNVs

Ref: Supplementary Information of [3]

After filtering properly mapped tags and then removing PCR duplication with SAMtools (ver. 0.1.19), we identified SNPs from the aligned sequence data using SAMtools Mpileup and BCFtools (ver. 0.1.19) with default options. We also identified reliable SNVs including heterozygous alleles from aligned sequence data that satisfied the following criteria using SAMtools Mpileup after filtering the properly mapped tags. (1) Sequence coverage > 20. (2) At least 25% of reads supported alternative alleles that differed from the B6 reference sequence.

flags in .sam/.bam file

$ samtools flags
Flags:
	0x1	PAIRED        .. paired-end (or multiple-segment) sequencing technology
	0x2	PROPER_PAIR   .. each segment properly aligned according to the aligner
	0x4	UNMAP         .. segment unmapped
	0x8	MUNMAP        .. next segment in the template unmapped
	0x10	REVERSE       .. SEQ is reverse complemented
	0x20	MREVERSE      .. SEQ of the next segment in the template is reversed
	0x40	READ1         .. the first segment in the template
	0x80	READ2         .. the last segment in the template
	0x100	SECONDARY     .. secondary alignment
	0x200	QCFAIL        .. not passing quality controls
	0x400	DUP           .. PCR or optical duplicate
	0x800	SUPPLEMENTARY .. supplementary alignment

0x 8 4 2 1 8 4 2 1 8 4 2 1
S d f s 2 1 R r U u P p
prorperly paired 0 1 1
with itself and mate mapped 0 0 1
singletons 1 0 1
Ref: http://www.htslib.org/doc/samtools.html

I extracted properly paired reads.

$ samtools view -b -f 0x02 DRR028646.bam > DRR028646.p.bam
took 1h; from DRR028646.bam (32GB), got DRR028646.p.bam (28GB)

For the removal of potential PCR duplicates, I used picard MarkDuplicates instead of samtools rmdup.

$ cd
$ wget -c https://github.com/broadinstitute/picard/releases/download/2.6.0/picard.jar
$ cd stap/reads/FES1
$ java -jar ~/picard.jar MarkDuplicates I=DRR028646.p.bam O=DRR028646.p.MD.bam M=DRR028646.picard_metrics.txt VALIDATION_STRINGENCY=SILENT
took 2h; got DRR028646.p.MD (29GB)

By using samtools mpileup, I got DRR028646.p.MD.vcf.gz. I used an automation script get_vcf2.sh in the real processings.

$ cd
$ nohup ./get_vcf2.sh DRR028646 &
$ nohup ./get_vcf2.sh DRR028647 &

"Reliable SNVs"

I obtained the list of reliable SNPs in FES1 data by using a script
extract-reliable-SNPs.py by Kanso

import sys

def main():
	# usage $python extract-reliable-SNPs.py [command]
	# ouput in VCF format
	min_allele_ratio = 0.25
	min_depth = 20+1
	is_skip_counts = False

	if len(sys.argv) == 2:
		if sys.argv[1] == "ignore-allele-ratio":
			min_allele_ratio = 0.0
		elif sys.argv[1] == "skip-counts":
			is_skip_counts = True
		else:
			print >> sys.stderr, "uknown command, neither of\n\tignore-allele-ratio\n\tskip-counts"
			return 3
	
	col_chr = 0
	col_pos = 1
	col_SNPid = 2
	col_ref = 3
	col_alt = 4
	col_qual = 5
	col_filter = 6
	col_info = 7
	
	lineno = 0
	
	with sys.stdin as fin: #use as a filter
		for line in fin:
			li = line.rstrip()
			if (len(li) >= 2) and (li[0] == '#') and (li[1] == '#'):
				continue
			ls = li.split("\t")
			lineno = lineno + 1
			if lineno == 1: #check file header
				labels = { col_chr:"#CHROM", col_pos:"POS", col_SNPid:"ID", col_ref:"REF", col_alt:"ALT", col_qual:"QUAL", col_filter:"FILTER", col_info:"INFO" }
				for col in labels.keys():
					if ls[col] != labels[col]:
						print >> sys.stderr, "column '" + labels[col] + "' could not be found in the input VCF file"
						return 2
				print "#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO"
			else:
				info = ls[col_info].split(";")
				depth = 0
				ref_count = 0
				alt_count = 0
				alt_allele_ratio = 0.0
				for i in info:
					if i == "INDEL":
						break
					key_value = i.split("=")
					if key_value[0] == "DP":
						depth = int(key_value[1])
					elif key_value[0] == "DP4":
						counts = key_value[1].split(",")
						ref_count = int(counts[0])+int(counts[1])
						alt_count = int(counts[2])+int(counts[3])
						if depth != 0:
							alt_allele_ratio = float(alt_count)/depth
				if (depth >= min_depth) and (alt_allele_ratio >= min_allele_ratio):
					#print li
					if is_skip_counts:
						print "%s\t%s\t%s\t%s\t%s\t.\t.\t." %(ls[0],ls[1],ls[2],ls[3],ls[4])
					else:	
						print "%s\t%s\t%s\t%s\t%s\t.\t.\t%d,%d" %(ls[0],ls[1],ls[2],ls[3],ls[4], ref_count,alt_count)
	return 0						

if __name__ == '__main__':
	main()
$ zcat DRR028646.p.MD.vcf.gz | python ~/extract-reliable-SNPs.py skip-counts > DRR028646.p.MD.erSNPs.vcf
$ zcat DRR028647.p.MD.vcf.gz | python ~/extract-reliable-SNPs.py skip-counts > DRR028647.p.MD.erSNPs.vcf
$ cat DRR028646.p.MD.erSNPs.vcf DRR028647.p.MD.erSNPs.vcf | LC_COLLATE=C sort -u > temp
$ grep '^#' temp > FES1.erSNPs.vcf && grep -v '^#' temp | LC_ALL=C sort -t '	' -k1,1 -k2,2n >> FES1.erSNPs.vcf
Note: Input ' ' as a single quote, CTRL-V and then TAB, a single quote in the command shell.

$ grep -v "^#" FES1.erSNPs.vcf | wc -l
780088

extracting reliable SNVs in FES1 that are not observed in FES2

From now on, I will use a bug-fixed version of snpexp (snpexp2). This fixed version is essential to obtain precise results in this section.

extract-negatives.py by Kanso

import sys, getopt

try:
	opts, args = getopt.getopt(sys.argv[1:], "m:")
except getopt.GetoptError as err:
	print str(err)
	sys.exit(1)

if len(args)!=1:
	sys.exit(2)

filename = args[0]
base_pos = { "A":0, "C":1, "G":2, "T":3 }
min_total_count = 1

for o,a in opts:
	if o=="-m":
		if a.isdigit():
			min_total_count = int(a)
		else:
			print >> sys.stderr, "argument for -m option should be non-negative integer"
			sys.exit(3)
	else:
		assert False, "undefined option"

with open(filename, "r") as fin:
	print "#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO"
	for line in fin:
		li = line.rstrip()
		ls = li.split("\t")
		alt = ls[5-1].split(",")
		acgt = ls[6-1].split(",")
		if len(acgt) == 4:
			alt_count = 0
			for a in alt:
				alt_count = alt_count + int(acgt[base_pos[a]])
			if alt_count == 0:
				total_count = 0
				for i in [0,1,2,3]:
					total_count = total_count + int(acgt[i])
				if total_count >= min_total_count:
					#print li
					print "%s\t%s\t%s\t%s\t%s\t.\t.\t." %(ls[0],ls[1],ls[2],ls[3],ls[4])

fix-und.sh by Kanso

#!/bin/bash
set -e
set -u
set -o pipefail

grep -v "undetermined" $1 > $1.temp
mv $1.temp $1

exit 0

$ ~/snpexp/src/snpexp2 -V ../FES1/FES1.erSNPs.vcf DRR028652.bam > DRR028652.FES1.erSNPs.snpexp2.out
$ ~/snpexp/src/snpexp2 -V ../FES1/FES1.erSNPs.vcf DRR028653.bam > DRR028653.FES1.erSNPs.snpexp2.out
$ ./fix-und.sh DRR028652.FES1.erSNPs.snpexp2.out
$ ./fix-und.sh DRR028653.FES1.erSNPs.snpexp2.out

$ python ~/extract-negatives.py DRR028652.FES1.erSNPs.snpexp2.out > DRR028652.-FES1.erSNPs.vcf
$ python ~/extract-negatives.py DRR028653.FES1.erSNPs.snpexp2.out > DRR028653.-FES1.erSNPs.vcf

$ comm -1 -2 <(sort DRR028652.-FES1.erSNPs.vcf) <(sort DRR028653.-FES1.erSNPs.vcf) > temp
$ grep "^#" temp > FES2-FES1.erSNPs.vcf; grep -v "^#" temp | LC_ALL=C sort -t '      ' -k1,1 -k2,2n >> FES2-FES1.erSNPs.vcf
fix-und.sh is required to remove trailing garbases in the outputs that are often added when snpexp causes segmentation fault. This process is not required after fixing this problem.
$ grep -v "^#" FES2-FES1.erSNPs.vcf | wc -l
$ cut -f1 FES2-FES1.erSNPs.vcf | uniq -c
14147

      1 #CHROM
     11 1
     11 10
   8889 11
   4541 12
     14 13
      1 14
      6 15
      6 16
     10 17
      7 18
      3 19
     12 2
      9 3
     14 4
     18 5
    567 6
     13 7
      7 8
      6 9
      1 X
      1 Y

reverse case: reliable SNVs in FES2 that are not observed in FES1

Starting from the extraction of reliable SNVs in FES2, I obtained unique SNVs for FES2 side.

$ grep -v "^#" FES2.erSNPs.vcf | wc -l
455928
$ grep -v "^#" FES1-FES2.erSNPs.vcf | wc -l
$ cut -f1 FES1-FES2.erSNPs.vcf | uniq -c
135

      1 #CHROM
     15 1
      7 10
      9 11
      8 12
     11 13
      8 14
      3 15
      1 16
      5 17
      3 18
      1 19
      9 2
      4 3
      3 4
     14 5
      4 6
      9 7
      8 8
      9 9
      4 Y

Presence of FES1/FES2-specific SNVs in the STAP-related cells

Note: I fixed an additional minor bug in snpexp. I used this fixed version in this section.

$ ~/snpexp/src/snpexp2 -V ../FES2/FES2-FES1.erSNPs.vcf DRR028632.bam > DRR028632.FES2-1.erSNPs.snpexp2.out &
$ ~/snpexp/src/snpexp2 -V ../FES1/FES1-FES2.erSNPs.vcf DRR028632.bam > DRR028632.FES1-2.erSNPs.snpexp2.out &

Extracting SNVs that exists in one of FES cell lines but not in the other can remove the effect of genetic background common to the parental mice used to establish these ES cells. But the genome-wide comparison between FES1 and FES2 SNPs distributions has reveled the presence of FES1-specific 129/129-homo clusters in some of the chromosomes [ref1,3]. This suggests that heterogeneity in the chromosomal level exists in the parental B6 mice and is differentially inherited in FES cell lines. To remove SNVs that are considered to be originated from this chromosomal event, I extracted SNVs in regions other than chromosome 6, 11, and 12 where the FES1-specific 129/129-home clusters are observed.

$ grep -v -E '^6|^11|^12' DRR028632.FES2-1.erSNPs.snpexp2.out > DRR028632.FES1-SNV.snpexp2.out
$ grep -v -E '^6|^11|^12' DRR028632.FES1-2.erSNPs.snpexp2.out > DRR028632.FES2-SNV.snpexp2.out
$ grep -v "^#" DRR028632.FES1-SNV.snpexp2.out | wc -l
$ grep -v "^#" DRR028632.FES2-SNV.snpexp2.out | wc -l
150
114

for FLS3 (rep1):

Summary for all the related cells (and a model):
full resolution

Data analysis environment

Can mutation accumulation explain genome difference between the STAP related cells and the ES cells?

...

Reference

  1. Katsura report (J)(E)
  2. RIKEN CDB press release 1/15/2015
  3. Nature B.C.A., "STAP cells are derived from ES cells"
  4. Supplements for the report from Review Committee for the Reconstruction of Management in RIKEN CDB