Authors: Don Conrad, Avinash Ramu, Kael Dai, and Reed A. Cartwright.
DeNovoGear is a software package to detect de novo mutations using next-generation sequencing data. It supports the analysis of many differential experimental designs and uses advanced statistical models to reduce the false positve rate.
Source code and binaries are available at https://github.com/denovogear/denovogear/releases.
Compilation of DeNovoGear requires CMake. Most Linux distributions allow you to install CMake using their package software.
tar -xvzf denovogear*.tar.gz
cd denovogear*/build
cmake ..
make
tar -xvzf denovogear*.tar.gz
cd denovogear*/build
cmake -DCMAKE_BUILD_TYPE=Release ..
make
tar -xvzf denovogear*.tar.gz
cd denovogear*/build
cmake -DBUILD_EXTERNAL_PROJECTS=1 ..
make
cd denovogear*/build
sudo make install
cd denovogear*/build
cmake -DCMAKE_INSTALL_PREFIX="${HOME}/dng"
make install
Denovogear comes with unit tests as well as full-suite test data available at https://github.com/denovogear/testdata. After running the build commands the tests can be downloaded and ran using:
make testdata
make test
dng call -p pedigree.ped seq1.bam seq2.bam seq3.bam ...
dng call -p pedigree.ped vars.bcf
dng dnm [auto|XS|SD] -p pedigree.ped -b vars.bcf
dng phase --dng dnm_variants --pgt parental_gts --bam seq.bam
dng call is a module that utilizes advanced peeling algorithms to identify mutations and other statistics on zero-loop pedigrees from SAM/BAM/CRAM files. dng call produces valid variant call format files (vcf or bcf) that can be used with other applications.
dng call is capable of calcuating:
- The probability of there being exactly one, or at least one, mutation at a site.
- The likely library genotype at a variant site, including quality score and posterior probabilities.
- The expected number of de novo mutations.
- The likelihood of observed data at a site.
- The Anderson-Darling test statistic for reference vs. alternative alleles.
- The likely gametic genotype of the parents in a trio.
Example: dng call -p family.ped familyL1.bam familyL2.bam familyL3.bam
Print Usage: dng help call
dng call uses a 6-column, tab-separated pedigree file format
1 1 0 0 1 NA12891
1 2 0 0 2 NA12892
1 3 1 2 2 NA12878
Column 1: Family ID
Column 2: Individual ID
Column 3: Father ID (0=unknown)
Column 4: Mother ID (0=unknown)
Column 5: Sex (1=male, 2=female, 0=unknown)
Column 6: Sample IDs; a newick-formatted tree
dng dnm takes in a PED file and a BCF file as input. The PED file describes the relationship between the samples and the BCF file contains the sequencing information for every locus.
`dng dnm auto --ped testdata/sample_CEU/sample_CEU.ped --bcf testdata/sample_CEU/sample_CEU.bcf`
If you would rather start with the BAM files and have samtools installed, this would work,
`samtools mpileup -gDf hg19.fa s1.bam s2.bam s3.bam | dng dnm auto --ped sample.ped --bcf -`
BCF files can be generated from the alignment using the samtools mpileup
command. The command to generate a bcf file from sample.bam is:
samtools mpileup -gDf reference.fa sample.bam > sample.bcf
The -D option of the samtools mpileup command retains the per-sample read depth
which is preferred by denovogear as it helps to filter out sites without a minimum number of reads(but note that DNG will work without per-sample RD information, in which case the RD tag encodes the average read depth information). The -g option computes genotype likelihoods and produces a compressed bcf output and the -f option is used to indicate the reference fasta file against which the alignment was built. A sample BCF file 'sample_CEU.bcf' is included in the distribution.
The PED file contains information about the trios present in the BCF file. Please make sure that all the members of the trios specified in the PED file are present in the BCF file. The PED file can be used to specify a subset of individuals for analysis from the BCF (in other words not every sample in the BCF needs to be represented in the PED file).
The PED file is a tab delimited file. The first six columns of the PED file are mandatory, these are Family ID, Individual ID, Paternal ID, Maternal ID, Sex (1 = male; 2 = female; other = unknown) and Phenotype. Denovogear makes use of the first four columns. The sample IDs in the PED file need to be exactly as they appear in the BCF file header. Sample order within the PED file does not matter, as family relationships are completely specified by the value of the child/mother/father fields in each row.
For example, a single line in the PED file that specifies a trio looks like:
CEU NA12878_vald-sorted.bam.bam NA12891_vald-sorted.bam.bam NA12892_vald-sorted.bam.bam 2 2
An example PED file, sample_CEU.ped, is included in the distribution directory.
These are tables with precomputed priors (and other useful numbers) for all possible trio configurations, under the null (no mutation present) and alternative (true de novo). The default tables are generated during each program run using a prior of 1 x 10 ^-8 /bp/generation on the haploid germline point mutation rate, and 1 x 10 ^-9 /bp/generation on the haploid germline indel mutation rate.
If you wish to change the default point or indel mutation rates use the --snp_mrate or --indel_mrate switches respectively.
For example
dng dnm auto --ped testdata/sample_CEU/sample.ped --bcf testdata/sample_CEU/sample.bcf --snp_mrate 2e-10 --indel_mrate 1e-11
[OR]
dng dnm auto --ped testdata/sample_CEU/sample.ped --vcf testdata/sample_CEU/sample.vcf --snp_mrate 2e-10 --indel_mrate 1e-11
The indel mutation rate prior is calculated based on the length of the insertion or deletion event, separate models are used for insertions and deletions. The two models are based on the indel observations from the 1000Genomes phase 1 data.
The insertion mutation rate is modeled using the function log (mrate) = mu_scale * (-22.8689 - (0.2994 * insertionLength))
The deletion mutation rate is modeled using the function log (mrate) = mu_scale * (-21.9313 - (0.2856 * deletionLength))
Note that a constant factor is used to scale the mutation rate, it is set to 1.0 by default and can be set using the switch --mu_scale. This provides the users to scale the mutation rate prior according to their data-set.
For example,
dng dnm auto --ped testdata/sample_CEU/sample.ped --bcf testdata/sample_CEU/sample.bcf --mu_scale 3
[OR]
dng dnm auto --ped testdata/sample_CEU/sample.ped --vcf testdata/sample_CEU/sample.vcf --mu_scale 3
The output format is a single row for each putative de novo mutation (DNM), with the following fields
1. Event type (POINT MUTATION or INDEL).
2. CHILD_ID - sample ID of trio-offspring with the DNM.
3. chr - chromosome.
4. pos - physical Position.
5. ref - base present in reference sequence at this position (from BCF).
6. alt - comma separated list of alternate non-reference alleles called on at-least one sample (from BCF).
7. maxlike_null - likelihood of the most likely mendelian-compatible genotype configuration.
8. pp_null - posterior probability of the most likely mendelian-compatible genotype configuration.
9. tgt_null(child/mom/dad) - genotypes of the most likely mendelian-compatible configuration.
10. snpcode - code that indicates whether the configuration shown in field 6 is monomorphic (1) or contains variation (2)(internal filters, can be ignored).
11. code - This field seems to be redundant to field 7, except the codes are (6) and (9).(internal filters, can be ignored).
12. maxlike_dnm - likelihood of the most likely DENOVO genotype configuration.
13. pp_dnm - posterior probability of the most likely DENOVO genotype configuration. Range = [0-1], a value closer to 1 indicates higher probability of observing a denovo event at this position. This is the field that is used to rank the calls.
14. tgt_dnm(child/mom/dad) - genotypes of the most likely mendelian-compatible configuration.
15. lookup - Code that indicates if the most likely DNM is a transition (4) or transversion (5) (for development use only).
16. flag - Flag that indicates whether the data for the site passed internal QC thresholds (for development use only).
17-19. Read depth of child, parent 1 and parent 2.
20-22. Root mean square of the mapping qualities of all reads mapping to the site for child, parent 1 and parent 2. Currently these values are the same for all samples when using BCF as the input format.
Fields 17-22 are meant for filtering out low quality sites.
Denovogear has separate models for autosomes, X chromosome in male offspring and X chromosome in female offspring. To use this model, create separate BCFs for the X chromosome.
`dng dnm auto --ped sample.ped --bcf sample.bcf`
[OR]
`dng dnm auto --ped sample.ped --vcf sample.vcf`
`dng dnm XS --ped sample.ped --bcf sample.bcf --region X`
[OR]
`dng dnm XS --ped sample.ped --vcf sample.X.vcf` (only BCF allows for random access)
`dng dnm XD --ped sample.ped --bcf sample.bcf --region X`
[OR]
`dng dnm XD --ped sample.ped --vcf sample.X.vcf`
DNG can be used to analyze paired samples for example to call somatic mutations between tumor/matched-normal pairs, the main difference in how to run the program is the way samples are specified in the PED file(see below),
`dng dnm auto --ped paired.ped --bcf sample.bcf`
[OR]
`dng dnm auto --ped paired.ped --vcf sample.vcf`
About the arguments,
1. paired.ped is a ped file containing the family-name and the name of the two samples in the bcf file. The last three columns are mandated by the PED format but are ignored by the program. An example line looks like
YRI NA19240_blood_vald-sorted.bam.bam NA19240_vald-sorted.bam.bam 0 0 0
2. sample.bcf[vcf] is a BCF[VCF] file containing both the samples.
A sample PED file sample_paired.ped for paired sample analysis is provided with the package.
The output format is a single row for each putative paired denovo mutation(DNM), with the following fields
1. Event type (POINT MUTATION or INDEL).
2. TUMOR_ID - sample ID of the 'TUMOR' sample.
3. NORMAL_ID - sample ID of the 'NORMAL' sample.
4. chr - chromosome.
5. pos - physical Position.
6. ref - base present in reference sequence at this position (from BCF).
7. alt - comma separated list of alternate non-reference alleles called on at-least one sample (from BCF).
8. maxlike_null - likelihood of the most likely compatible genotype configuration.
9. pp_null - posterior probability of the most likely compatible genotype configuration.
10. tgt_null(normal/tumor) - genotypes of the most likely compatible configuration.
11. maxlike_dnm - likelihood of the most likely denovo genotype configuration.
12. pp_dnm - posterior probability of the most likely denovo genotype configuration, a value closer to 1 indicates strong evidence of a denovo event. This is the field that is used to rank the calls.
13. tgt_dnm(normal/tumor) - genotypes of the most likely denovo configuration.
14-15. Read depth of tumor, normal samples.
16-17. Root mean square of the mapping qualities of all reads mapping to the site for tumor, normal samples.
18-19. null_snpcode, dnm_snpcode - snpcode is a field used to classify the genotype configurations for the null and alternate case. 1 stands for hom/hom, 2 stands for het/hom, 3 stands for hom/het and 4 stands for het/het.
DNG can be used to obtain parental phasing information for Denovo Mutations where phase informative sites are present. This is done by looking at reads which cover both the denovo base and a phase informative positions. Phase informative positions are SNP positions that lie within a certain window from the denovo site, the default window size is 1000 bp but the window-size can be set by the user. For each DNM, a list of phase informative sites from the genotypes file is obtained, these are the loci which lie within the phasing window. Also, these sites do not have a het/het GT configuration for the parents and where the child is het. The number of DNM and phasing allele combinations seen in the read level data is output by the program. For the phasing sites, the inferred parental origin is displayed. For example if the base at the phasing site is T and the parental genotypes are P1:CC and p2:TC at this site, then the parent of origin of this base is p2. By looking at the base in the denovo position on this read it is possible to infer the parent of origin of the denovo mutation.
A sample list of dnms file, sample_phasing_dnm_f is provided. Also a sample genotypes file sample_phasing_GTs_f is provided for reference.
`dng phaser --dnm dnm_f --pgt gts_file --bam alignment.bam --window 1000`
About the arguments,
1. "dnm" is the list of DNMs whose parental origin is to be determined. It is a tab delimited file of the format
chr pos inherited_base mutant_base
2. "pgt" contains the genotypes of the child and the parents at SNP sites. It is a tab delimited file of the format
chr pos child_GT parent1_GT parent2_GT
3. "bam" is the alignment file (.bam) of the child.
4. "window" is an optional argument which is the maximum distance between the DNM and a phasing site. The default value is 1000.
DNM_pos 1:182974758 INHERITED G VARIANT A
HAP POS 182974328 p1: CC p2: TC Number of denovo-phasing pairs found: 0
HAP POS 182974572 p1: CC p2: TC Number of denovo-phasing pairs found: 2
Base at DNM position: A Base at phasing position: C INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 3
Base at DNM position: G Base at phasing position: T INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 3
HAP POS 182974598 p1: GG p2: CG Number of denovo-phasing pairs found: 2
Base at DNM position: A Base at phasing position: G INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 1
Base at DNM position: G Base at phasing position: C INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 2
HAP POS 182974602 p1: TT p2: CT Number of denovo-phasing pairs found: 2
Base at DNM position: A Base at phasing position: T INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 1
Base at DNM position: G Base at phasing position: C INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 1
HAP POS 182974707 p1: TT p2: CT Number of denovo-phasing pairs found: 2
Base at DNM position: A Base at phasing position: T INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 25
Base at DNM position: G Base at phasing position: C INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 24
HAP POS 182974750 p1: AA p2: TA Number of denovo-phasing pairs found: 2
Base at DNM position: A Base at phasing position: A INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 45
Base at DNM position: G Base at phasing position: T INFERRED PARENT OF ORIGIN for DNM: p1 SUPPORTING READ COUNT: 47
Please feel free to contact the authors about any concerns/comments.
###General options for TRIOs and Paired Sample Calling
--snp_mrate: Mutation rate prior for SNPs. [1e-8] --indel_mrate: Mutation rate prior for INDELs. [1e-9] --pair_mrate: Mutation rate prior for paired sample analysis. [1e-9] --indel_mu_scale: Scaling factor for indel mutation rate. [1] --pp_cutoff: Posterior probability threshold. [0.0001] --rd_cutoff: Read depth filter, sites where either one of the sample have read depth less than this threshold are filtered out. [10 --region: region of the BCF file over which to perform denovo calling. [string of the form "chr:start-end"]
- FEATURE: When building with cmake, users have the option to download and install dependencies.
Just run
cmake -DBUILD_EXTERNAL_PROJECTS=1 .. - FEATURE:
dng helpnow fully implemented - FEATURE: more information added to
dng calloutput - BUGFIX: When running
dng callwith HTSLIB 1.2.1, the following error message was being emitted: "FIXME: dirty header not synced". This has been fixed. - BUGFIX:
dng callnow outputs correct 1-based site locations. - BUGFIX: a segfault was fixed in
dng dnmanddng phasercaused by invalid commandline arguments - CHANGE: HTSLIB 1.2+ is now required.
- Miscellaneous improvements to the build system
- Main program now called 'dng'
- Added experimental 'dng call' module.
- DeNovoGear now requires HTSLIB 1+, CMake 3.1+, Boost 1.47+, and Eigen 3+.
- made changes to indel_mrate parameter
- better indenting
- mu_scale scales indel mutation rate linearly
- added GPL v3
- updated output fields for indels, snps to be the same
- removed 'X' allele in VCF op. VCF can be indexed by Tabix, IGVTools and used in Annovar.
- added region based denovo calling on BCF files, invoked with --region flag
- added vcf parser for denovo calling, invoked with --vcf flag
- Added read-depth, posterior-probability filters.
- Output number of sites in the BCF and number of sites passing filters.
- Modified paired caller output.
- Fixed bug in triallelic configuration.
- Some trialleic denovo configurations were being called incorrectly.
- Recent C++ compiler, supporting C++11 (eg. gcc 4.8.1+ or clang 3.3+)
- CMake 3.1+ when compiling http://www.cmake.org/download/#latest
- HTSlib 1.2+ http://www.htslib.org/
- Eigen 3 http://eigen.tuxfamily.org/
- Boost 1.47+ http://www.boost.org/
Most Unix distributions contain package software that will install these dependencies for you. DNG contains code to download and build missing dependencies.
dng treecall is an experimental module to identify mutations and trees from whole-genome sequencing of single cells
Print Usage: dng treecall -h
see https://github.com/nh3/treecall/
DeNovoGear uses the Samtools libraries to parse BCF files, we thank the Heng Li and the rest of the authors of Samtools for making this resource available to developers.