Germline variant detection

Based on the read mapping, germline variants are identified at positions where the read alignment supports a significant difference to the reference genome.

This is achieved through a site model, where each position is first assigned a likelihood for each of the genotypes A, C, T, G, N or missing. The algorithm then iterates over the read mapping and adjusts likelihoods per position for each genotype based on observations in the data until the likelihoods no longer change. Note that broken read pairs are not considered.

Each position is then inspected, and positions where the most likely genotype(s) are different from the reference sequence are identified.

At this stage, homopolymer variants with a homopolymer length of >=5 are re-called. This is done by calculating the likelihood of all possible genotypes based on the homopolymer length variants found in the reads, with the assumption that all other homopolymer variants in the reads arise by error. The likelihood is (up to a normalizing constant):

$\displaystyle \prod_{j=1}^{n} (\sum_{i} P(l_j \vert l_i) f_j) ^{c_j} $

where $ P(l_j \vert l_i)$ is the probability of observing a homopolymer of length $ l_j$ by error when the true length is $ l_i$, and $ c_j$ is the number of fragments with a homopolymer of length $ l_j$. $ f_j$ is the frequency of the homopolymer with length $ l_j$ according to the genotype $ G$. For example, for diploid models this frequency can be 0, 0.5, or 1.0. The probabilities $ P(l_j \vert l_i)$ are determined from the sample, by counting the number of homopolymer errors at positions that appear to be homozygous.

The final homopolymer variants are those that maximize the likelihood. However, if the maximum likelihood genotype is nearly homozygous (by which we mean all except one haplotype has the same variant), then we perform an additional test to see whether a ploidy-0.1:0.1 frequency ratio between the two variants has higher likelihood than the ploidy-1:1 frequency ratio. If it does, then we call the variant as homozygous. This ensures that low levels of noise are tolerated, and improves the accuracy of homopolymer calls.

Notes

Special handling is applied to variants supported by only 1 read that have a coverage of 1 or 2. For details, see the description of the Allele count option under Variant filters in LightSpeed Fastq to Germline Variants.

For insertions only, unaligned ends that are shorter than the full insertion, but matches the insertion sequence, contribute to the count and coverage.

A limit of maximum three alleles is enforced for each homopolymer locus and for alleles specifically marked with STR "Yes" that affect the same short tandem repeat. The alleles with the highest read counts are retained. See the description of the STR annotations and filter option under Variant filters in LightSpeed Fastq to Germline Variants for details about STR annotation.

When enabling the option use non-specific reads for variant detection, for sites with at least 80% ambiguous reads, the sensitivity to heterozygous events is increased. The reason is, that the non-specific reads often spread variant alleles across two or more similar sites, resulting in alleles with lower than 50% allele frequency at the individual sites.

Variant types

LightSpeed Fastq to Germline Variants reports SNPs, MNVs and InDels and replacements provided that the variants are contained within at least one paired end read.

Variant annotations

Variants identified by LightSpeed Fastq to Germline Variants are annotated with the following basic information: Chromosome, Region, Type, Reference, Allele, Reference allele, Length, Zygosity, Count, Coverage, Frequency, QUAL and Genotype. Only single base pair variants, that are not adjacent to any other variants, are assigned a QUAL score.

Read about general variant annotations here: https://resources.qiagenbioinformatics.com/manuals/clcgenomicsworkbench/current/index.php?manual=Variant_tracks.html.

In addition to the basic annotations, a number of LightSpeed specific annotations are available:

Note that for insertions, counts from unaligned ends that are shorter than the full insertion, but matches the insertion sequence, are included in the variant annotations Count, Coverage, Frequency, Count (singleton UMI), Count (big UMI), and Proportion (singleton UMIs). Counts from unaligned ends are not included in Forward read count, Reverse read count, Forward coverage, reverse coverage and Forward/reverse balance.