We will illustrate how TrioCaller works in sequence data including trios and unrelated samples. We will walk through all necessary steps to move from raw sequence data to called genotypes. If you are new to sequence data, please review every step. If you are experienced, you may directly jump to TrioCaller specific section.
We will start with a set of short sequence reads and associated base quality scores (stored in a fastq file), find the most likely genomic location for each read (producing a BAM file), generate an initial list of polymorphic sites and genotypes (stored in a VCF file) and use haplotype information to refine these genotypes (resulting in an updated VCF file).
If you are interested in de novo mutations or are working on one or two families with deep sequence data (>30X), you should first consider our sister program, Polymutt, which ignores linkage disequilibrium information but can handle more complex pedigrees.
Before downloading the program, we appreciate if you could email email@example.com (Subject: TrioCaller, with/without a little descriptive information (e.g. Affiliation, depth, the number of samples and family structure). We will notify you if there is any update.
A recent extension of TrioCaller: FamLDCaller is coming soon with major updates (better processing function, handling general families and reference panels). Please try the beta version below. Contact firstname.lastname@example.org for any questions.
An example from sequence data to genotypes
The example dataset demonstrated here is also included. Our dataset consists of 40 individuals, including 10 parent-offspring trios and 10 unrelated individuals. The average sequence depth is ~3x. README.txt describes the structure of the package. Pipeline.csh (C shell) and pipeline.bash (bash shell) are two scripts for you to run all commands listed here in batch.
To conserve time and disk-space, our analysis will focus on a small region on chromosome 20 around position 2,000,000. We will first map reads for a single individual (labeled SAMPLE1). Then we combine the results with mapped reads from all individuals to generate a list of polymorphic sites and estimate accurate genotypes at each of these sites.
In addition to TrioCaller, you will need BWA (available from Sourceforge) and samtools (also from Sourceforge) installed to run this exercise. The examples are tested in in bwa 0.6.1, samtools 0.1.18, TrioCaller 0.1.1; we expect newer versions should also work. We assume all executables are in your path.
Building an Index for Short Read Alignment
To quickly place short reads along the genome, BWA and other read mappers typically build a word index for the genome. This index lists the location of particular short words along the genome and can be used to seed and then extend particular matches.
The sequence index is typically not compatible across different BWA versions. To rebuild the sequence index, issue the following commands:
#Remove any earlier reference files rm ref/human_g1k_v37_chr20.fa.* #Rebuild the reference bin/bwa index -a is ref/human_g1k_v37_chr20.fa
Mapping Reads to The Genome
Next, we will use BWA to find the most likely sequence location for each read using the
bwa aln command. This command requires two parameters, one corresponding to the reference genome, the other corresponding to a fastq file containing reads to be mapped.
bin/bwa aln -q 15 ref/human_g1k_v37_chr20.fa fastq/SAMPLE1.fastq > bwa.sai/SAMPLE1.sai
The file SAMPLE1.fastq contains DNA sequence reads for sample SAMPLE1.
A fastq file consists of a series of multi-line records. Each record starts with a read name, followed by a DNA sequencing, a separator line, and a set of per base quality scores. Base quality scores estimate the probability of error at each sequenced base (a base quality of 10 denotes an error probability of 10%, base quality 20 denotes 1% error probability and base quality 30 denotes 0.1% error probability). These error probabilities are each encoded in a single character for compactness and can be decoded using an ASCII table (simply look up the ascii code for each base and subtract 33 to get base quality). By inspecting the FastQ file you should be able to learn about the length of reads being mapped and their base qualities. For example, try to figure out if base quality is typically higher at the start or end of each read...
Converting Alignments to BAM format
The .sai alignment format is specific to BWA, so the first thing to do is to convert the alignment to a more standard format that will be compatible with downstream analysis tools. We can do this with a combination of the
bwa samse command and
samtools view and
samtoosl sort commands.
bin/bwa samse -r "@RG\tID:ILLUMINA\tSM:SAMPLE1" ref/human_g1k_v37_chr20.fa bwa.sai/SAMPLE1.sai fastq/SAMPLE1.fastq | \ bin/samtools view -uhS - | bin/samtools sort -m 2000000000 - bams/SAMPLE1
You can check the use of parameters in the bwa manual. The result BAM file uses a compact binary format to represent the
alignment of each short read to the genome. You can view the contents
of the file using the
samtools view command, like so:
bin/samtools view bams/SAMPLE1.bam | more
The text representation of the alignment produced by
samtools view describes
the alignment of one read per line. The most interesting fields are column 1 (the read
name), columns 3 and 4 (the alignment position), column 5 (the CIGAR string, describing
any gaps in the alignment), and columns 10 and 11 (with the sequence and quality score). In this representation, all alignments are automatically converted to the forward strand.
Indexing the BAM file
If you reached this far, rejoice! The mapping process is almost done. We will now create
an index for the file, which makes it convenient to quickly extract reads from any
genome location. We do this with the
samtools index command, like so:
bin/samtools index bams/SAMPLE1.bam
Browsing Alignment Results
You can now view the contents of the alignment at any location using the
samtools tview commands. While the
tview generates prettier output,
it is not compatible with all screens. For example, to view reads overlapping
starting at position 2,100,000 on chromosome 20, we could run:
bin/samtools tview bams/SAMPLE1.bam ref/human_g1k_v37_chr20.fa
Then, type "g 20:2100000"
So let's recap: we have mapped reads to genome, converted them from a BWA specific format to a more commonly used format used by many different programs, sorted and indexed the results. In most cases, the next step would be to remove duplicate reads and to ensure that base quality scores are properly calibrated. To save time, we'll skip those steps now.
Till now, we only finished read mapping for one sample SAMPLE1. We need to repeat this step for other samples (SAMPL2 - SAMPLE40). You can try something like:
For c shell
foreach file (`ls fastq/SAMPLE*.fastq | cut -f 2 -d '/' | cut -f 1 -d '.'`) echo $file bin/bwa aln -q 15 ref/human_g1k_v37_chr20.fa fastq/$file.fastq > bwa.sai/$file.sai bin/bwa samse -r "@RG\tID:ILLUMINA\tSM:$file" ref/human_g1k_v37_chr20.fa bwa.sai/$file.sai fastq/$file.fastq | \ bin/samtools view -uhS - | bin/samtools sort -m 2000000000 - bams/$file bin/samtools index bams/$file.bam end
For bash shell
for file in `ls fastq/SAMPLE*.fastq | cut -f 2 -d '/' | cut -f 1 -d '.'`; do echo $file; bin/bwa aln -q 15 ref/human_g1k_v37_chr20.fa fastq/$file.fastq > bwa.sai/$file.sai; bin/bwa samse -r "@RG\tID:ILLUMINA\tSM:$file" ref/human_g1k_v37_chr20.fa bwa.sai/$file.sai fastq/$file.fastq | \ bin/samtools view -uhS - | samtools sort -m 2000000000 - bams/$file; bin/samtools index bams/$file.bam; done
Once we finish the read mapping step and generate bam files for all samples, we can step to variant calling and genotype inference.
Calling variants and Inferring genotypes
Initial set of variant calls
You probably thought the initial mapping process was quite convoluted ... you'll be glad to know that the next few steps are much simpler.
The first thing we'll do is use samtools to generate an initial list of variant sites, using the
mpileup command. This command looks at the bases aligned to each location and flags locations that are likely to vary. By default, the results are stored in BCF file, which can be converted into the more widely used VCF format using bcftools (a companion set of tools distributed with samtools).
bin/samtools mpileup -Iuf ref/human_g1k_v37_chr20.fa bams/SAMPLE*bam | bin/bcftools view -bvcg - > result/chr20.mpileup.bcf
bin/bcftools view result/chr20.mpileup.bcf > result/chr20.mpileup.vcf
The VCF format is a simple text format. It starts with several header lines, which all start with the two '##' characters, and is followed by a single line per marker that provides both summary information about the marker and genotypes for each individual. You can review the contents of the VCF file using the 'more' command:
Here are some questions for you to investigate:
- How many variant sites were detected in this dataset? Try a command like this one:
grep -vE ^# result/chr20.mpileup.vcf | wc -l
(The grep command line excludes all lines beginning with # and then the wc command counts the number of lines in the file).
Genotype Refinement Using Linkage Disequilibrium Information (TrioCaller)
The initial set of genotype calls is generated examining a single individual at a time. These calls are typically quite good for deep sequencing data, but much less accurate for low pass sequence data. In either case, they can be greatly improved by models that combine information across sites and individuals and consider the contraints imposed by parent-offspring trios.
Note: The current version only supports SNP data, so please filter indels before running TrioCaller. It supports VCF 4.0 and 4.1 formats with the exception of dropped missing trailing fields (e.g. use complete missing notation ./.:.:.:.,.,. rather than ./. for the genotype field)
Here is a summary of the TrioCaller command line options (these are also listed whenever you run the program with no parameters):
Available Options Shotgun Sequences: --vcf , --pedfile  Markov Sampler: --seed , --burnin , --rounds  Haplotyper: --states , --errorRate , --compact Phasing: --randomPhase , --inputPhased, --refPhased Output Files: --prefix , --phase, --interimInterval  Explanation of Options --vcf: Standard VCF file (4.0 and above). --pedfile: Pedigree file in MERLIN format. --seed: Seed for sampling, default 123456. --burnin: The number of rounds ignored at the beginning of sampling. --rounds: The total number of iterations. --states: The number of haplotyes used in the state space. The default is the maximum number: 2*(number of founders -1). --errorRate: The pre-defined base error rate. Default 0.01. --randomPhase: The initial haplotypes are inferred from the single marker. Default option. --inputPhased: The initial haplotypes are directly from input VCF file (with "|" as separator in the genotype column). --refPhased: The initial haplotypes are built on reference alleles from VCF file. --prefix: The prefix of output file --interimInterval: The number of rounds for interim outputs
Note: The pedigree files require complete trio structures (all three members of the trio exist in the file). For parent-offspring pair, create a "fake" parent in the pedigree file or code them as unrelated individuals. The order of the names in the pedigree file is NOT necessary to be consistent with that in .vcf file. The computation will be intensive if the number of samples are large. You can use option "--states" to reduce the computation cost (e.g. start with "--states 50")
To complete our example analysis, we could run:
bin/TrioCaller --vcf result/chr20.mpileup.vcf --pedfile ped/triocaller.ped --states 50 --rounds 10 --prefix result/chr20.triocaller
The format of output file is same as the input file. Again, you can review the contents of the updated VCF file using the more command:
All right. Congratulations! You have come to the end and learned basic skills for accurate genotype calling in trios.
Chen W, Li B, Zeng Z, Sanna S, Sidore C, Busonero F, Kang HM, Li Y, Abecasis GR. Genotype calling and haplotyping in parent-offspring trios. Genome Res. 2013 Jan;23(1):142-51 LINK