Monopogen: SNV calling from single cell sequencing data

News

4/12/2024: Version 1.6.0 released.
- We added the guide on putative SNV filtering based on cell type information derived from single cell RNA/ATAC-seq modalites.
2/26/2024: Version 1.5.0 released.
- In the cell-scan step, we implemented a motif-based search on wild/mutated alleles for all cells from the bam file directly. The single-cell level bam file splitting and joint calling modules were removed. This new version achieves over 10-fold speed up than the old version due to avoid the bam splitting. It could take less than 60 mins to collect the wild/mutated allele profiles of 10K cells over 20K loci.
- Recommended hard-filterings on putative somatic SNVs from Monopogen were added.

Introduction

Monopogen is an analysis package for SNV calling from single-cell sequencing, developed and maintained by Ken chen's lab in MDACC. Monopogen works on sequencing datasets generated from single cell RNA 10x 5', 10x 3', single ATAC-seq technoloiges, scDNA-seq etc.

It is composed of three modules:

Data preprocess. This module removes reads with high alignment mismatches from single cell sequencing and also makes data formats compatiable with Monopongen.
Germline SNV calling. Given the sparsity of single cell sequencing data, we leverage linkage disequilibrium (LD) from external reference panel(such as 1KG3, TopMed) to improve both SNV calling accuracy and detection sensitivity.
Putative somatic SNV calling. We extended the machinery of LD refinement from human population level to cell population level. We statistically phase the observed alleles with adjacent germline alleles to estimate the degree of LD, taking into consideration widespread sparseness and allelic dropout in single-cell sequencing data, and calculated a probabilistic score as an indicator of somatic SNVs.

The output of Monopogen will enable 1) ancestry identificaiton on single cell samples; 2) genome-wide association study on the celluar level if sample size is sufficient, and 3) putative somatic SNV investigation.

Installation

Dependencies

python (version >= 3.73)
java (open JDK>=1.8.0)
R (version >= 4.0.0)
pandas>=1.2.3
pysam>=0.16.0.1
NumPy>=1.19.5
sciPy>=1.6.3
pillow>=8.2.0
data.table(R package; version >=1.14.8)
e1071 (R package; 1.7-13)
ggplot2

!Note We have put the binary compatibility tools including samtools, bcftools, beagle in the app folder. We fixed the version because the output formats vary a lot with different versions. If you are not able to run them, you can compile them in you system. We only test on these tools on following versions:

samtools Version: 1.2 (using htslib 1.2.1)
bcftools Version: 1.8 (using htslib 1.8)
beagle.27Jul16.86a.jar (version 4.1)
tabix Version: 1.9
bgzip Version: 1.9

If you meet other errors when running Monopogen, go to FAQs section.

Installation

Right now Monopogen is avaiable on github, you can install it through github

git clone https://github.com/KChen-lab/Monopogen.git
cd Monopogen
pip install -e .

Quick Start

Note the quick start exmaple only works for germline module. If you want to test somatic module, please go the section

Somatic SNV calling from scRNA-seq

For quick start of Monopogen, we provide an example dataset provided the example/ folder, which includes:

A.bam (.bai)
The bam file storing read alignment for sample A.
B.bam (.bai)
The bam file storing read alignment for sample B.
CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
The reference panel with over 3,000 samples in 1000 Genome database. Only SNVs located in chr20: 0-2Mb were extracted in this vcf file.
chr20_2Mb.hg38.fa (.fai)
The genome reference used for read aligments. Only seuqences in chr20:0-2Mb were extracted in this fasta file. There are three test scripts in the test/ folder test/runPreprocess.sh, test/runGermline.sh, test/runSomatic.sh for quick start of Monopogen

Data preprocess

You can type the following command to get the help information.

path="XXX/Monopogen"  # where Monopogen is downloaded
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${path}/apps
python ${path}/src/Monopogen.py  preProcess --help`

Output is

usage: Monopogen.py preProcess [-h] -b BAMFILE [-o OUT] -a APP_PATH
                               [-m MAX_MISMATCH] [-t NTHREADS]

optional arguments:
  -h, --help            show this help message and exit
  -b BAMFILE, --bamFile BAMFILE
                        The bam file for the study sample, the bam file should
                        be sorted. If there are multiple samples, each row
                        with each sample (default: None)
  -o OUT, --out OUT     The output director (default: None)
  -a APP_PATH, --app-path APP_PATH
                        The app library paths used in the tool (default: None)
  -m MAX_MISMATCH, --max-mismatch MAX_MISMATCH
                        The maximal alignment mismatch allowed in one reads
                        for variant calling (default: 3)
  -t NTHREADS, --nthreads NTHREADS
                        Number of threads used for SNVs calling (default: 1)

You need to prepare the bam file list for option -b.

path="XXy/Monopogen"
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${path}/apps
python  ${path}/src/Monopogen.py  preProcess -b bam.lst -o out  -a ${path}/apps

After running the preProcess module, there will be bam files after quality controls in the folder out/Bam/ used for downstream SNV calling.

Germline SNV calling

You can type the following command to get the help information.

python ${path}/src/Monopogen.py  germline --help`

The output is

usage: Monopogen.py germline [-h] -r REGION -s
                             {varScan,varImpute,varPhasing,all} [-o OUT] -g
                             REFERENCE -p IMPUTATION_PANEL
                             [-m MAX_SOFTCLIPPED] -a APP_PATH [-t NTHREADS]

optional arguments:
  -h, --help            show this help message and exit
  -r REGION, --region REGION
                        The genome regions for variant calling (default: None)
  -s {varScan,varImpute,varPhasing,all}, --step {varScan,varImpute,varPhasing,all}
                        Run germline variant calling step by step (default:
                        all)
  -o OUT, --out OUT     The output director (default: None)
  -g REFERENCE, --reference REFERENCE
                        The human genome reference used for alignment
                        (default: None)
  -p IMPUTATION_PANEL, --imputation-panel IMPUTATION_PANEL
                        The population-level variant panel for variant
                        imputation refinement, such as 1000 Genome 3 (default:
                        None)
  -a APP_PATH, --app-path APP_PATH
                        The app library paths used in the tool (default: None)
  -t NTHREADS, --nthreads NTHREADS
                        Number of threads used for SNVs calling (default: 1)

You need to prepare the genome region file list for option -r with an example shown in test/region.lst. We also included an optimal genome region file in ${path}/resource/GRCh38.region.lst for the whole genome SNV calling. Each region is in one row. Run the test script test/runGermline.sh as following:

python  ${path}/src/Monopogen.py  germline  \
    -a   ${path}/apps -t 1   -r  region.lst \
    -p  ../example/  \
    -g  ../example/chr20_2Mb.hg38.fa   -s all  -o out

The germline module will generate the phased VCF files with name *.phased.vcf.gz in the folder out/germline. If there are multiple samples in the bam file list from -b option in preProcess module, the phased VCF files will contain genotypes from multiple samples. The output of phased genotypes are as following:

##fileformat=VCFv4.2
##filedate=20240227
##source="beagle.27Jul16.86a.jar (version 4.1)"
##INFO=<ID=AF,Number=A,Type=Float,Description="Estimated ALT Allele Frequencies">
##INFO=<ID=AR2,Number=1,Type=Float,Description="Allelic R-Squared: estimated squared correlation
##INFO=<ID=DR2,Number=1,Type=Float,Description="Dosage R-Squared: estimated squared correlation
##INFO=<ID=IMP,Number=0,Type=Flag,Description="Imputed marker">
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
##FORMAT=<ID=DS,Number=A,Type=Float,Description="estimated ALT dose [P(RA) + P(AA)]">
##FORMAT=<ID=GP,Number=G,Type=Float,Description="Estimated Genotype Probability">
#CHROM  POS     ID      REF     ALT     QUAL    FILTER  INFO    FORMAT  A       B
chr20   60291   .       G       T       .       PASS    .       GT      0|1     0|0
chr20   63117   .       T       C       .       PASS    .       GT      0|0     1|0
chr20   64506   .       C       T       .       PASS    .       GT      0|0     0|0
chr20   68303   .       T       C       .       PASS    .       GT      0|1     1|1
chr20   75250   .       C       T       .       PASS    .       GT      0|1     0|0
chr20   88108   .       T       C       .       PASS    .       GT      1|1     1|0
chr20   101433  .       A       C       .       PASS    .       GT      0|0     0|1
chr20   101498  .       A       G       .       PASS    .       GT      0|0     1|1
chr20   127687  .       A       C       .       PASS    .       GT      1|1     1|1
chr20   140857  .       C       A       .       PASS    .       GT      0|0     0|1
chr20   153835  .       T       C       .       PASS    .       GT      0|1     1|1
chr20   154002  .       C       T       .       PASS    .       GT      1|1     1|1
chr20   159104  .       T       C       .       PASS    .       GT      1|1     1|1
chr20   165212  .       C       A       .       PASS    .       GT      0|0     1|1
chr20   167839  .       T       C       .       PASS    .       GT      1|1     1|1
chr20   175269  .       T       C       .       PASS    .       GT      1|1     0|0
chr20   186086  .       G       A       .       PASS    .       GT      1|1     0|0
chr20   186183  .       G       A       .       PASS    .       GT      1|1     0|0

Run on the HPC

If there are multiple single cell RNA samples and you want to use Monopogen on germline SNV calling, you can enable the -norun option.

python  ${path}/src/Monopogen.py  germline  \
    -a   ${path}/apps -t 8   -r  region.lst \
    -p  ../example/  \
    -g  ../example/chr20_2Mb.hg38.fa   -s all  -o out
    --norun TRUE

The germline outputs for the demo data could be seen in test/chr20.gl.vcf.gz, test/chr20.gp.vcf.gz and test/chr20.phased.vcf.gz. The -norun module will generate jobs from different regions and you can submit them to HPC based on your own preference. The generated job files will be in out/Script/

Germline SNV calling from snRNA-seq

We demonstrate the utilization of Monopogen on germline SNV calling, ancestry identification on snRNA samples from human retina atlas. The 4 retina samples shown in Monopogen methodological paper are 19D013, 19D014, 19D015, 19D016. Thhe fastq files of these samples can be downloaded with from SRA database SRR23617370, SRR23617337, SRR23617320 and SRR23617310. Here we used 19D013 as an example (analysis on other samples is the same).

variant calling

For convenience, we skipped the read alignment step and shared the alignmed bam file (Only reads from chr20 were extracted) in 19D013.snRNA.chr20.bam and 19D013.snRNA.chr20.bam.bai. Users also need to prepare for the GRCh38 reference (with chr as prefix in the sequence ID) used for read alignment and 1KG3 imputation panel from 1KG3. We can prepare for the bam.lst and region.lst as following

less bam.lst

The output is

19D013,19D013.snRNA.chr20.bam

less region.lst

The output is

chr20

Please make sure all required files available

ls
19D013.snRNA.chr20.bam
19D013.snRNA.chr20.bam.bai
bam.lst
CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
GRCh38.chr20.fa
region.lst

The data preprocess step can be run as (~3 mins)

path="/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen"
${path}/src/Monopogen.py  preProcess -b bam.lst -o retina  -a ${path}/apps  -t 1

The output is

[2023-04-25 16:23:05,747] INFO     Monopogen.py Performing data preprocess before variant calling...
[2023-04-25 16:23:05,747] INFO     Monopogen.py Parameters in effect:
[2023-04-25 16:23:05,748] INFO     Monopogen.py --subcommand = [preProcess]
[2023-04-25 16:23:05,748] INFO     Monopogen.py --bamFile = [bam.lst]
[2023-04-25 16:23:05,748] INFO     Monopogen.py --out = [retina]
[2023-04-25 16:23:05,748] INFO     Monopogen.py --app_path = [/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps]
[2023-04-25 16:23:05,748] INFO     Monopogen.py --max_mismatch = [3]
[2023-04-25 16:23:05,748] INFO     Monopogen.py --nthreads = [1]
[2023-04-25 16:23:05,765] DEBUG    Monopogen.py PreProcessing sample 19D013
[2023-04-25 16:25:56,543] INFO     Monopogen.py Success! See instructions above.

The germline SNV calling can be run as (~80 mins).

${path}/src/Monopogen.py  germline  -a ${path}/apps  -r region.lst \
 -p ./ \
 -g  GRCh38.chr20.fa  -m 3 -s all  -o retina

The output is

[2023-04-25 16:30:39,749] INFO     Monopogen.py Performing germline variant calling...
[2023-04-25 16:30:39,749] INFO     Monopogen.py Parameters in effect:
[2023-04-25 16:30:39,749] INFO     Monopogen.py --subcommand = [germline]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --region = [region.lst]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --step = [all]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --out = [retina]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --reference = [GRCh38.chr20.fa]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --imputation_panel = [CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --max_softClipped = [3]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --app_path = [/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps]
[2023-04-25 16:30:39,749] INFO     Monopogen.py --nthreads = [1]
[2023-04-25 16:30:39,750] INFO     Monopogen.py Checking existence of essenstial resource files...
[2023-04-25 16:30:39,754] INFO     Monopogen.py Checking dependencies...
['bash retina/Script/runGermline_chr20.sh']
[fai_load] build FASTA index.
[mpileup] 1 samples in 1 input files
(mpileup) Max depth is above 1M. Potential memory hog!
Lines   total/split/realigned/skipped:  56054517/437864/36916/0
beagle.27Jul16.86a.jar (version 4.1)
Copyright (C) 2014-2015 Brian L. Browning
Enter "java -jar beagle.27Jul16.86a.jar" for a summary of command line arguments.
Start time: 05:10 PM CDT on 25 Apr 2023

Command line: java -Xmx18204m -jar beagle.jar
  gl=retina/germline/chr20.gl.vcf.gz
  ref=CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
  chrom=chr20
  out=retina/germline/chr20.gp
  impute=false
  modelscale=2
  nthreads=1
  gprobs=true
  niterations=0

No genetic map is specified: using 1 cM = 1 Mb

reference samples:    3202
target samples:          1

Window 1 [ chr20:60291-64332055 ]
reference markers:   31534
target markers:      31531
...
Number of reference markers:     31534
Number of target markers:        31531
Total time for building model: 22 minutes 17 seconds
Total time for sampling:       5 minutes 4 seconds
Total run time:                29 minutes 44 seconds

End time: 05:40 PM CDT on 25 Apr 2023
beagle.27Jul16.86a.jar (version 4.1) finished
beagle.27Jul16.86a.jar (version 4.1)
Copyright (C) 2014-2015 Brian L. Browning
Enter "java -jar beagle.27Jul16.86a.jar" for a summary of command line arguments.
Start time: 05:40 PM CDT on 25 Apr 2023

Command line: java -Xmx18204m -jar beagle.jar
  gt=retina/germline/chr20.germline.vcf
  ref=CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
  chrom=chr20
  out=retina/germline/chr20.phased
  impute=false
  modelscale=2
  nthreads=48
  gprobs=true
  niterations=0

No genetic map is specified: using 1 cM = 1 Mb

reference samples:    3202
target samples:          1

Window 1 [ chr20:60291-64331516 ]
reference markers:   23755
target markers:      23755

Starting burn-in iterations

Window=1 Iteration=1
Time for building model:         1 minute 29 seconds
Time for sampling (singles):     0 seconds
DAG statistics
mean edges/level: 51     max edges/level: 122
mean edges/node:  1.206  mean count/edge: 126
...
Number of markers:               23755
Total time for building model: 14 minutes 0 seconds
Total time for sampling:       2 seconds
Total run time:                15 minutes 7 seconds

End time: 05:55 PM CDT on 25 Apr 2023
beagle.27Jul16.86a.jar (version 4.1) finished
[2023-04-25 17:55:37,771] INFO     Monopogen.py Success! See instructions above.

The final output of germline SNVs from Monopogen are in the folder retina/germline/chr20.phased.vcf.gz. These phased genotypes could be used for downstream ancestry identification, association study, and somatic SNV calling.

##fileformat=VCFv4.2
##filedate=20230425
##source="beagle.27Jul16.86a.jar (version 4.1)"
##INFO=<ID=AF,Number=A,Type=Float,Description="Estimated ALT Allele Frequencies">
##INFO=<ID=AR2,Number=1,Type=Float,Description="Allelic R-Squared: estimated squared correlation b
##INFO=<ID=DR2,Number=1,Type=Float,Description="Dosage R-Squared: estimated squared correlation be
##INFO=<ID=IMP,Number=0,Type=Flag,Description="Imputed marker">
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
##FORMAT=<ID=DS,Number=A,Type=Float,Description="estimated ALT dose [P(RA) + P(AA)]">
##FORMAT=<ID=GP,Number=G,Type=Float,Description="Estimated Genotype Probability">
#CHROM  POS     ID      REF     ALT     QUAL    FILTER  INFO    FORMAT  19D013_European_F_78
chr20   60291   .       G       T       .       PASS    .       GT      1|0
chr20   68303   .       T       C       .       PASS    .       GT      1|0
chr20   75250   .       C       T       .       PASS    .       GT      1|0
chr20   88108   .       T       C       .       PASS    .       GT      1|1
chr20   101574  .       G       A       .       PASS    .       GT      1|0
chr20   101576  .       G       A       .       PASS    .       GT      1|1
chr20   159104  .       T       C       .       PASS    .       GT      1|1
chr20   175269  .       T       C       .       PASS    .       GT      1|1
chr20   186086  .       G       A       .       PASS    .       GT      1|1
chr20   186183  .       G       A       .       PASS    .       GT      1|1
chr20   198814  .       A       T       .       PASS    .       GT      1|0
chr20   203580  .       G       A       .       PASS    .       GT      1|1
chr20   213223  .       G       C       .       PASS    .       GT      0|1
chr20   213244  .       A       G       .       PASS    .       GT      0|1
chr20   231710  .       T       G       .       PASS    .       GT      1|1

genotyping accuracy evaluation

We can validate the genotyping accuracy and sensitvity (recall) by comparing Monopogen outputs with matched WGS-based genotypes. Users can download the WGS-based genotypes from chr22 only 19D013.wgs.chr20.vcf. We use vcftools to compare genotypes of Monopogen to the gold standard. Before evaluation, you need to remove the homozygous included in the phasing results.

zless ./retina/germline/chr20.phased.vcf.gz | grep -v "0|0" | bgzip -c > ./retina/germline/chr20.phased.het.vcf.gz 
vcftools --gzvcf  ./retina/germline/chr20.phased.het.vcf.gz    --diff  19D013.wgs.chr20.vcf   --diff-discordance-matrix --out  19D013  --chr chr20

The output is

VCFtools - 0.1.15
(C) Adam Auton and Anthony Marcketta 2009

Parameters as interpreted:
        --gzvcf ./retina/germline/chr20.phased.vcf.gz
        --chr chr20
        --out 19D013
        --diff 19D013.wgs.chr20.vcf
        --diff-discordance-matrix

Using zlib version: 1.2.3
Versions of zlib >= 1.2.4 will be *much* faster when reading zipped VCF files.
After filtering, kept 1 out of 1 Individuals
Outputting Discordance Matrix
        For bi-allelic loci, called in both files, with matching alleles only...
Non-matching ALT. Skipping all such sites.
Non-matching REF. Skipping all such sites.
Found 23290 sites common to both files.
Found 464 sites only in main file.
Found 85853 sites only in second file.
After filtering, kept 23755 out of a possible 23755 Sites
Run Time = 0.00 seconds

Monopogen can detect 21.3% (23290/(23290+85853)) germline SNVs although the singel cell data is quite sparisty. Remarkably, the false positive rate is lower than 2% (464/(464+23290)). The genotype concordance could be further examined based on the overlapped 23290 SNVs by looking at the output of 19D013.diff.discordance_matrix.

less 19D013.diff.discordance_matrix
-       N_0/0_file1     N_0/1_file1     N_1/1_file1     N_./._file1
N_0/0_file2     0       0       0       0
N_0/1_file2     0       13628   723     0
N_1/1_file2     0       60      8869    0
N_./._file2     0       0       0       0

The genotyping concordance is calculated as 97% ((60+723)/(60+723+13628+8869)). The overall genotyping accuracy could be 95% (0.97*(1-0.02))

ancestry identification

Here we demonstrate how we can identify ancestry background on snRNA sample 19D013 based on the output of Monopogen. Users can use LASER/TRACE software to project 19D013 on HGDP reference panel. The HGDP genotyping panel was already included in the LASER/TRACE software. Before that, we need to liftover Monopogen output from GRCh38 to GRCh37 to match the HGDP genotyping coordinates. The fasta file of GRCh37 on chr20 could be downloaded as GRCh37.chr20.fa.

chain="${path}/resource/hg38ToHg19.over.chain.gz"
GRCh37_chr20="GRCh37.chr20.fa"
picard="${path}/apps/picard.jar"

java -jar ${picard} CreateSequenceDictionary R=${GRCh37_chr20} O="GRCh37.chr20.dict"
java -Xmx10g  -jar ${picard} LiftoverVcf I=./retina/germline/chr20.phased.vcf.gz    O=./retina/germline/chr20.phased.GRCh37.vcf.gz   R=${GRCh37_chr20}  CHAIN=${chain} REJECT="temp.vcf"   WARN_ON_MISSING_CONTIG=true

The output will be as following

INFO    2023-04-26 01:45:14     CreateSequenceDictionary

********** NOTE: Picard's command line syntax is changing.
**********
********** For more information, please see:
********** https://github.com/broadinstitute/picard/wiki/Command-Line-Syntax-Transition-For-Users-(Pre-Transition)
**********
********** The command line looks like this in the new syntax:
**********
**********    CreateSequenceDictionary -R GRCh37.chr20.fa -O GRCh37.chr20.dict
**********

01:45:14.905 INFO  NativeLibraryLoader - Loading libgkl_compression.so from jar:file:/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps/picard.jar!/com/intel/gkl/native/libgkl_compression.so
[Wed Apr 26 01:45:14 CDT 2023] CreateSequenceDictionary OUTPUT=GRCh37.chr20.dict REFERENCE=GRCh37.chr20.fa    TRUNCATE_NAMES_AT_WHITESPACE=true NUM_SEQUENCES=2147483647 VERBOSITY=INFO QUIET=false VALIDATION_STRINGENCY=STRICT COMPRESSION_LEVEL=5 MAX_RECORDS_IN_RAM=500000 CREATE_INDEX=false CREATE_MD5_FILE=false GA4GH_CLIENT_SECRETS=client_secrets.json USE_JDK_DEFLATER=false USE_JDK_INFLATER=false
[Wed Apr 26 01:45:14 CDT 2023] Executing as jdou1@ldragon2 on Linux 3.10.0-1160.15.2.el7.x86_64 amd64; OpenJDK 64-Bit Server VM 1.8.0_312-b07; Deflater: Intel; Inflater: Intel; Provider GCS is not available; Picard version: 2.26.10
[Wed Apr 26 01:45:14 CDT 2023] picard.sam.CreateSequenceDictionary done. Elapsed time: 0.00 minutes.
Runtime.totalMemory()=2058354688
To get help, see http://broadinstitute.github.io/picard/index.html#GettingHelp
Exception in thread "main" picard.PicardException: /rsrch3/scratch/bcb/jdou1/scAncestry/retina/bam_backup/monopogen_demo1/GRCh37.chr20.dict already exists.  Delete this file and try again, or specify a different output file.
        at picard.sam.CreateSequenceDictionary.doWork(CreateSequenceDictionary.java:220)
        at picard.cmdline.CommandLineProgram.instanceMain(CommandLineProgram.java:308)
        at picard.cmdline.PicardCommandLine.instanceMain(PicardCommandLine.java:103)
        at picard.cmdline.PicardCommandLine.main(PicardCommandLine.java:113)
INFO    2023-04-26 01:45:15     LiftoverVcf

********** NOTE: Picard's command line syntax is changing.
**********
********** For more information, please see:
********** https://github.com/broadinstitute/picard/wiki/Command-Line-Syntax-Transition-For-Users-(Pre-Transition)
**********
********** The command line looks like this in the new syntax:
**********
**********    LiftoverVcf -I ./retina/germline/chr20.phased.vcf.gz -O ./retina/germline/chr20.phased.GRCh37.vcf.gz -R GRCh37.chr20.fa -CHAIN /rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/resource/hg38ToHg19.over.chain.gz -REJECT temp.vcf -WARN_ON_MISSING_CONTIG true
**********


01:45:15.530 INFO  NativeLibraryLoader - Loading libgkl_compression.so from jar:file:/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps/picard.jar!/com/intel/gkl/native/libgkl_compression.so
[Wed Apr 26 01:45:15 CDT 2023] LiftoverVcf INPUT=./retina/germline/chr20.phased.vcf.gz OUTPUT=./retina/germline/chr20.phased.GRCh37.vcf.gz CHAIN=/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/resource/hg38ToHg19.over.chain.gz REJECT=temp.vcf WARN_ON_MISSING_CONTIG=true REFERENCE_SEQUENCE=GRCh37.chr20.fa    LOG_FAILED_INTERVALS=true WRITE_ORIGINAL_POSITION=false WRITE_ORIGINAL_ALLELES=false LIFTOVER_MIN_MATCH=1.0 ALLOW_MISSING_FIELDS_IN_HEADER=false RECOVER_SWAPPED_REF_ALT=false TAGS_TO_REVERSE=[AF] TAGS_TO_DROP=[MAX_AF] DISABLE_SORT=false VERBOSITY=INFO QUIET=false VALIDATION_STRINGENCY=STRICT COMPRESSION_LEVEL=5 MAX_RECORDS_IN_RAM=500000 CREATE_INDEX=false CREATE_MD5_FILE=false GA4GH_CLIENT_SECRETS=client_secrets.json USE_JDK_DEFLATER=false USE_JDK_INFLATER=false
[Wed Apr 26 01:45:15 CDT 2023] Executing as jdou1@ldragon2 on Linux 3.10.0-1160.15.2.el7.x86_64 amd64; OpenJDK 64-Bit Server VM 1.8.0_312-b07; Deflater: Intel; Inflater: Intel; Provider GCS is not available; Picard version: 2.26.10
INFO    2023-04-26 01:45:15     LiftoverVcf     Loading up the target reference genome.
INFO    2023-04-26 01:45:16     LiftoverVcf     Lifting variants over and sorting (not yet writing the output file.)
INFO    2023-04-26 01:45:16     LiftoverVcf     Processed 23755 variants.
INFO    2023-04-26 01:45:16     LiftoverVcf     184 variants failed to liftover.
INFO    2023-04-26 01:45:16     LiftoverVcf     99 variants lifted over but had mismatching reference alleles after lift over.
INFO    2023-04-26 01:45:16     LiftoverVcf     1.1913% of variants were not successfully lifted over and written to the output.
INFO    2023-04-26 01:45:16     LiftoverVcf     liftover success by source contig:
INFO    2023-04-26 01:45:16     LiftoverVcf     chr20: 23472 / 23755 (98.8087%)
INFO    2023-04-26 01:45:16     LiftoverVcf     lifted variants by target contig:
INFO    2023-04-26 01:45:16     LiftoverVcf     chr20: 23472
WARNING 2023-04-26 01:45:16     LiftoverVcf     99 variants with a swapped REF/ALT were identified, but were not recovered.  See RECOVER_SWAPPED_REF_ALT and associated caveats.
INFO    2023-04-26 01:45:16     LiftoverVcf     Writing out sorted records to final VCF.
[Wed Apr 26 01:45:16 CDT 2023] picard.vcf.LiftoverVcf done. Elapsed time: 0.02 minutes.
Runtime.totalMemory()=2058354688

Given SNVs from chr20 only are not enough to identify individual ancestry, we provided the VCF files 19D013.phased.GRCh37.vcf.gz by mering all 22 chromosomes. Users can run other chromosome using the same way as we did . Then we can run TRACE to project 19D013 on HGDP panel

zless -S ./retina/germline/19D013.phased.GRCh37.vcf.gz  | awk '{gsub(/\chr/, "")}1'  > 19D013.trace.vcf
/rsrch1/bcb/kchen_group/ytan1/LASER-2.04/vcf2geno/vcf2geno --inVcf 19D013.trace.vcf  --out 19D013.trace
/rsrch1/bcb/kchen_group/ytan1/LASER-2.04/trace  -s 19D013.trace.geno  \
  -g /rsrch1/bcb/kchen_group/ytan1/LASER-2.04/HGDP/HGDP_938.geno \
  -c /rsrch1/bcb/kchen_group/ytan1/LASER-2.04/HGDP/HGDP_938.RefPC.coord  \

The output is

Analysis started at: Wed Apr 26 02:33:45 2023
The following parameters are available.  Ones with "[]" are in effect:

Available Options
           Input/Output : --inVcf [19D013.trace.vcf], --out [19D013.trace]
          People Filter : --peopleIncludeID [], --peopleIncludeFile []
                          --peopleExcludeID [], --peopleExcludeFile []
            Site Filter : --rangeList [], --rangeFile []
      Auxilary Function : --keepDuplication, --updateID []
...
Skip duplicated variant site:  [ 18     56371446        .       C       G ]
Skip duplicated variant site:  [ 18     77922913        .       A       C ]
Skip duplicated variant site:  [ 19     1489460 .       C       T ]
Skip duplicated variant site:  [ 22     50273174        .       A       C ]
Total 830699 VCF records have converted successfully
Total 1 people and 830652 markers are outputted

=====================================================================
====    TRACE: fasT and Robust Ancestry Coordinate Estimation    ====
====          Version 1.03, Last updated on Dec/30/2016          ====
====          (C) 2013-2016 Chaolong Wang, GNU GPL v3.0          ====
=====================================================================
Started at: Wed Apr 26 02:33:48 2023

1 individuals are detected in the STUDY_FILE.
830652 loci are detected in the STUDY_FILE.
938 individuals are detected in the GENO_FILE.
Warning: Two datasets have different alleles at locus [8:2929436]: [A,G] vs [A,T].
Warning: Two datasets have different alleles at locus [12:5734319]: [A,G] vs [A,C].
Warning: Two datasets have different alleles at locus [13:109351901]: [T,G] vs [T,A].
632958 loci are detected in the GENO_FILE.
938 individuals are detected in the COORD_FILE.
100 PCs are detected in the COORD_FILE.

Parameter values used in execution:
-------------------------------------------------
STUDY_FILE (-s) 19D013.trace.geno
GENO_FILE (-g)  /rsrch1/bcb/kchen_group/ytan1/LASER-2.04/HGDP/HGDP_938.geno
COORD_FILE (-c) /rsrch1/bcb/kchen_group/ytan1/LASER-2.04/HGDP/HGDP_938.RefPC.coord
OUT_PREFIX (-o) trace
DIM (-k)        2
DIM_HIGH (-K)   20
THRESHOLD (-t)  1e-06
MIN_LOCI (-l)   100
FIRST_IND (-x)  1
LAST_IND (-y)   1
KNN_ZSCORE (-knn)       10
RANDOM_SEED (-seed)     0
NUM_THREADS (-nt)       8
-------------------------------------------------

Wed Apr 26 02:33:54 2023
Identify 85497 loci shared by STUDY_FILE and GENO_FILE.
Exclude 3 loci that have different alleles in two datasets.
The analysis will base on the remaining 85494 shared loci.

Wed Apr 26 02:33:54 2023
Reading reference genotype data ...

Wed Apr 26 02:35:05 2023
Calculating reference covariance matrix ...

Wed Apr 26 02:35:06 2023
Reading reference PCA coordinates ...

Wed Apr 26 02:35:06 2023
Analyzing study individuals ...
Procrustean PCA coordinates are output to 'trace.ProPC.coord'.

Finished at: Wed Apr 26 02:35:06 2023
=====================================================================

The PCA coordinates of 19D013 is in the file trace.ProPC.coord.

less trace.ProPC.coord

popID   indivID L       K       t       Z       PC1     PC2
19D013_European_F_78    19D013_European_F_78    2546    20      0.98445 7.45623 93.869  164.372

We can show it on the HGDP PCA plot as

Rscript ${path}/resource/plotTrace.R  ${path}/resource/HGDP.PC.csv  trace.ProPC.coord 19D013_onHGDP

The PCA projection plot will be generated as 19D013_onHGDP.pdf

Somatic SNV calling from scRNA-seq

We demonstrate how the LD refinement model implemented in Monopogen can improve somatic SNV detection from scRNA-seq profiles without matched bulk WGS data available. We used the benchmarking dataset of bone marrow single cell samples from Miller et al.,. The raw fastq files could be downloaded from SRA database with SRR15598778, SRR15598779, SRR15598780, SRR15598781, and SRR15598782. For convenience, we shared with the the downloaded bam file from chromosome 20 chr20.master_scRNA.bam.

preprocess

To remove reads with high alignment mismatches, we first run the preprocess step by setting the bam file list bam.lst as

less bam.lst
bm,chr20.maester_scRNA.bam

The data preprocess can be run as (~3 mins)

path="/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen"
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${path}/apps
python  ${path}/src/Monopogen.py  preProcess -b bam.lst -o bm  -a ${path}/apps -t 1

The output could be

[2023-05-07 08:37:50,307] INFO     Monopogen.py Performing data preprocess before variant calling...
[2023-05-07 08:37:50,307] INFO     germline.py Parameters in effect:
[2023-05-07 08:37:50,307] INFO     germline.py --subcommand = [preProcess]
[2023-05-07 08:37:50,307] INFO     germline.py --bamFile = [bam.lst]
[2023-05-07 08:37:50,307] INFO     germline.py --out = [bm]
[2023-05-07 08:37:50,307] INFO     germline.py --app_path = [/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps]
[2023-05-07 08:37:50,307] INFO     germline.py --max_mismatch = [3]
[2023-05-07 08:37:50,307] INFO     germline.py --nthreads = [1]
[2023-05-07 08:37:50,336] DEBUG    Monopogen.py PreProcessing sample bm
[2023-05-07 08:40:36,538] INFO     Monopogen.py Success! See instructions above.

germline calling

To detect putative somatic SNVs, we need to call germline module to build the LD refinement model. The required region.lst could be set as (Note, only the whole chromosome calling is allowed!)

less region.lst
chr20

Users also need to preprare for following files CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz from 1KG3 imputation panel from 1KG3 and GRCh38.chr20.fa.

${path}/src/Monopogen.py  germline  -a ${path}/apps  -r region.lst \
 -p ./  -t 22 \
 -g  GRCh38.chr20.fa  -m 3 -s all  -o bm

This will take ~ 25 mins with output as

[2023-05-07 09:25:43,724] INFO     Monopogen.py Performing germline variant calling...
[2023-05-07 09:25:43,724] INFO     germline.py Parameters in effect:
[2023-05-07 09:25:43,724] INFO     germline.py --subcommand = [germline]
[2023-05-07 09:25:43,724] INFO     germline.py --region = [region.lst]
[2023-05-07 09:25:43,724] INFO     germline.py --step = [all]
[2023-05-07 09:25:43,724] INFO     germline.py --out = [bm]
[2023-05-07 09:25:43,724] INFO     germline.py --reference = [GRCh38.chr20.fa]
[2023-05-07 09:25:43,724] INFO     germline.py --imputation_panel = [./]
[2023-05-07 09:25:43,724] INFO     germline.py --max_softClipped = [3]
[2023-05-07 09:25:43,724] INFO     germline.py --app_path = [/rsrch3/scratch/bcb/jdou1/scAncestry/Monopogen/apps]
[2023-05-07 09:25:43,724] INFO     germline.py --nthreads = [1]
[2023-05-07 09:25:43,724] INFO     germline.py --norun = [FALSE]
[2023-05-07 09:25:43,724] INFO     Monopogen.py Checking existence of essenstial resource files...
[2023-05-07 09:25:43,777] INFO     Monopogen.py Checking dependencies...
['bash bm/Script/runGermline_chr20.sh']
[mpileup] 1 samples in 1 input files
(mpileup) Max depth is above 1M. Potential memory hog!
Lines   total/split/realigned/skipped:  10933032/105880/21378/0
beagle.27Jul16.86a.jar (version 4.1)
Copyright (C) 2014-2015 Brian L. Browning
Enter "java -jar beagle.27Jul16.86a.jar" for a summary of command line arguments.
Start time: 09:33 AM CDT on 07 May 2023

Command line: java -Xmx18204m -jar beagle.jar
  gl=bm/germline/chr20.gl.vcf.gz
  ref=./CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
  chrom=chr20
  out=bm/germline/chr20.gp
  impute=false
  modelscale=2
  nthreads=1
  gprobs=true
  niterations=0

No genetic map is specified: using 1 cM = 1 Mb

reference samples:    3202
target samples:          1

Window 1 [ chr20:273372-39851321 ]
reference markers:   10486
target markers:      10486

Starting burn-in iterations

Window=1 Iteration=1
Time for building model:         39 seconds
Time for sampling (singles):     7 seconds
DAG statistics
mean edges/level: 49     max edges/level: 129
mean edges/node:  1.183  mean count/edge: 131
...
Number of markers:               10486
Total time for building model: 6 minutes 31 seconds
Total time for sampling:       1 minute 30 seconds
Total run time:                10 minutes 19 seconds

End time: 09:43 AM CDT on 07 May 2023
beagle.27Jul16.86a.jar (version 4.1) finished
beagle.27Jul16.86a.jar (version 4.1)
Copyright (C) 2014-2015 Brian L. Browning
Enter "java -jar beagle.27Jul16.86a.jar" for a summary of command line arguments.
Start time: 09:43 AM CDT on 07 May 2023

Command line: java -Xmx18204m -jar beagle.jar
  gt=bm/germline/chr20.germline.vcf
  ref=./CCDG_14151_B01_GRM_WGS_2020-08-05_chr20.filtered.shapeit2-duohmm-phased.vcf.gz
  chrom=chr20
  out=bm/germline/chr20.phased
  impute=false
  modelscale=2
  nthreads=1
  gprobs=true
  niterations=0

No genetic map is specified: using 1 cM = 1 Mb

reference samples:    3202
target samples:          1

Window 1 [ chr20:273372-39851321 ]
reference markers:    9130
target markers:       9130

Starting burn-in iterations

Window=1 Iteration=1
Time for building model:         28 seconds
Time for sampling (singles):     0 seconds
DAG statistics
mean edges/level: 48     max edges/level: 123
mean edges/node:  1.203  mean count/edge: 133
...
Number of markers:                9130
Total time for building model: 5 minutes 3 seconds
Total time for sampling:       1 second
Total run time:                6 minutes 59 seconds

End time: 09:50 AM CDT on 07 May 2023
beagle.27Jul16.86a.jar (version 4.1) finished
[2023-05-07 09:50:21,243] INFO     Monopogen.py Success! See instructions above.

ld refinement on putative somatic SNVs

One advantage of Monopogen is to extend the machinery of LD refinement from human population level to cell population level. Users need to prepare for the cell barcode file CB_7K.maester_scRNA.csv. The cell barcode file includes two column: 1) cell barcode; 2 number of reads detected in each cell. This could be from cell ranger/Seurat. Make sure the column names are cell and id in the cell barcode csv file. You can select top cells (1K~10K) with high reads detected. There are three steps featureInfo, cellScan, and LDrefinement to call putative somatic SNVs. Here we show the step one by one.

To extract the feature information from sequencing data, we need to run (this step will take ~22s). Note, the option -t enables users to run mulitple chromosomes simultaneously. Set -t=1 if you are working on only one chromosome.

python  ${path}/src/Monopogen.py  somatic  \
    -a   ${path}/apps  -r  region.lst  -t 1 \
    -i  bm  -l  CB_7K.maester_scRNA.csv   -s featureInfo     \
    -g   GRCh38.chr20.fa

The output would be

[2024-03-04 09:55:20,598] INFO     Monopogen.py Get feature information from sequencing data...
[2024-03-04 09:55:42,232] INFO     Monopogen.py Success! See instructions above.

Then, we need to collect single cell level read information by running the cellScan module as

python  ${path}/src/Monopogen.py  somatic  \
    -a   ${path}/apps  -r  region.lst  -t 1  \
    -i  bm  -l  CB_7K.maester_scRNA.csv   -s cellScan     \
    -g   GRCh38.chr20.fa

This process would take ~15 mins to be finished

[2024-03-04 09:55:42,651] INFO     Monopogen.py Collect single cell level information from sequencing data...
scanning read 1000000
scanning read 2000000
[2024-03-04 10:10:17,343] INFO     Monopogen.py Success! See instructions above.

Finally, we can run the LD refinment step to further improve the putative somatic SNV detection as (taking ~3 mins)

python  ${path}/src/Monopogen.py  somatic  \
    -a   ${path}/apps  -r  region.lst  -t 1 \
    -i  bm  -l  CB_7K.maester_scRNA.csv   -s LDrefinement     \
    -g   GRCh38.chr20.fa

After running the LDrefinment step, there would be two files chr20.germlineTwoLoci_model.csv andchr20.germlineTrioLoci_model.csv in the output directory bm/somatic. These two enable us to examine the rationale of the LD model in sparse data at the cell population level. Users can examine this by looking at output figure LDrefinement_germline.chr20.pdf

Users need to perform hard filtering based on the file chr20.putativeSNVs.csv as following

SVM_pos_score>0.5. The SVM_pos_score is the prediction score from the SVM module. Closing to 0 has higher probability of sequencing error.
LDrefine_merged_score>0.25. The LDrefine_merged_score is from the LDrefinement module. Closing to 0 is germline SNVs and closing to 0.5 is more likely the putative somatic SNVs. The NA values in LDrefine_merged_score column denotes that there are no informative germline SNVs tagging the putative somatic SNVs.
0.1<BAF_alt<0.5, Dep_ref>5, and Dep_alt>5. The BAF_alt is frequency of alternative allele, Dep_ref denotes the number of cells with only reference allele detected and Dep_alt for alternative allele.
remove germline SNVs overlapped in genomeAD database.

Users can also extract the reads covering putative SNVs at the single cell resolution from chr20.SNV_mat.RDS. Starting from column 19, each column denotes one cell. In each element (for example 1/0), the number denotes whether there is read supporting reference/alternative allele.

R
> dt < - readRDS(file="chr20.SNV_mat.RDS")
> dt[,seq(19,21,1))]
                 ATGACCAGTCACTAGT ACCCTTGGTCTCACAA TCCTCCCCAATACCTG
chr20:276310:A:G              0/0              0/0              0/0
chr20:391901:A:G              0/0              0/0              0/0
chr20:410498:T:C              1/0              0/0              0/0
chr20:410520:A:T              1/0              0/0              0/0
chr20:436781:A:G              0/0              0/0              0/0

Putative SNV filtering based on cell type information

More details could be see in following vignette

Putative SNV filtering based on cell type information

FAQs

Is Monopogen call SNVs from mitochondria genome?
No. Monopogen needs the LD from 1KG3 as input. Also, Monopogen does not work on mouse genome.
How to use the multi-threading function -t in Monopogen
With the putative somatic SNV calling, user can set -t as the number of chromosomes listed in region file
where to download 1KG3 reference panel (hg38) http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/data_collections/1000G_2504_high_coverage/working/20201028_3202_phased/
how to perform downstream PCA-based projection or admixture analysis
PCA-based projection analysis can be peformed using LASER 2.0
bcftools: error while loading shared libraries: libbz2.so.1.0: not able to open shared object file: No such file or directly
Adding the apps folder of Monopogen in your library environment

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/xx/apps
AssertionError: Program vcftools cannot be found!
You may set the read/write permission on the folder xx/apps as

chmod 770 -R /xx/apps

Citation

Dou J, Tan Y, Kock KH, Wang J, Cheng X, Tan LM, Han KY, Hon CC, Park WY, Shin JW, Jin H, H Chen, L Ding, S Prabhakar, N Navin. K Chen. Single-nucleotide variant calling in single-cell sequencing data with Monopogen. Nature Biotechnology. 2023 Aug 17:1-0

Name		Name	Last commit message	Last commit date
Latest commit History 299 Commits
Monopogen.egg-info		Monopogen.egg-info
apps		apps
example		example
resource		resource
src		src
test		test
LICENSE		LICENSE
README.md		README.md
requirements.txt		requirements.txt
setup.py		setup.py

License

KChen-lab/Monopogen

Folders and files

Latest commit

History

Repository files navigation

Monopogen: SNV calling from single cell sequencing data

News

Table of Contents

Introduction

Installation

Quick Start

Data preprocess

Germline SNV calling

Run on the HPC

Germline SNV calling from snRNA-seq

variant calling

genotyping accuracy evaluation

ancestry identification

Somatic SNV calling from scRNA-seq

preprocess

germline calling

ld refinement on putative somatic SNVs

Putative SNV filtering based on cell type information

FAQs

Citation

About

Topics

Resources

License

Stars

Watchers

Forks

Languages