Genome Sequencing Analysis: Variant Calling to Interpretation

Genome sequencing analysis turns the raw DNA reads from a sequencer into a filtered, annotated catalogue of variants and a biological interpretation of what they mean. The instrument produces billions of DNA letters in FASTQ files; whole genome sequencing analysis is everything that follows, from aligning those reads to a reference through calling and filtering variants to annotating their effect. This guide walks the pipeline for both eukaryotic and bacterial genomes and explains where the decisions that determine a trustworthy result actually sit.

If you have whole genome sequencing data and need variants called and annotated, or a set of bacterial isolates turned into a phylogenetic tree, this guide shows what a complete DNA sequencing analysis involves and how to keep it reproducible and defensible.

What Genome Analysis Is Trying to Find

A genome is mostly identical between any two individuals of a species; the interesting biology lives in the differences. Genome sequence analysis is therefore largely the business of finding and interpreting those differences, the variants: single-base substitutions, small insertions and deletions, and larger structural changes. For a human study, that means cataloguing how a sample's genome departs from the reference and predicting which of those departures matter. For a set of bacterial isolates, it means comparing genomes to each other to reconstruct how strains are related and what resistance or virulence genes they carry.

Because the reference build and annotation databases define the coordinate system and the meaning of every variant, a variant call is only interpretable alongside the exact versions used. Pinning and documenting them is not bureaucratic; it is what makes the result reproducible at all.

The Whole Genome Sequencing Pipeline

The reference-based pipeline for human and other eukaryotic genomes is well established and runs through these stages.

1. Quality control and trimming. Assess every FASTQ file with FastQC and MultiQC, then trim adapters and low-quality bases.
2. Alignment. Map reads to the reference genome with BWA-MEM, producing a sorted, indexed BAM file.
3. Preprocessing. Mark duplicate reads and, where appropriate, recalibrate base qualities so the variant caller is not misled by systematic errors.
4. Variant calling. Call single-nucleotide variants and small insertions and deletions with GATK or bcftools.
5. Filtering. Apply quality and depth filters, because raw variant calls are full of artifacts and an unfiltered call set is not a result.
6. Annotation. Predict the functional consequences of each variant with SnpEff or VEP and intersect with relevant databases.

Coverage matters throughout: roughly 30x is standard for germline whole genome sequencing, while somatic or low-frequency variant detection needs substantially more depth. For the statistical side of association testing or cohort comparison, our biostatistics support handles the modeling that sits on top of a called variant set.

Bacterial WGS, Assembly, and Phylogenetics

Bacterial whole genome sequencing analysis follows a different path from the human pipeline. Genomes are small, and the questions are usually comparative: how are these isolates related, which carry resistance genes, and where did an outbreak spread. The analysis can run through reference mapping and variant detection or through de novo genome assembly, which reconstructs each genome from its reads without leaning on a reference, the better choice for novel or highly divergent strains.

From there, the comparative workflow typically maps reads and detects variants across all isolates, removes recombinant regions (recombination violates the assumptions of standard phylogenetic models and must be filtered before tree-building), builds a maximum-likelihood phylogeny, and annotates the tree. Strain typing and resistance or virulence gene detection round out the picture. This full bacterial pipeline, version-pinned and delivered with the command log, is part of our omics analysis service.

DNA, RNA, and Choosing the Right Sequencing Question

Genome sequencing answers questions about the DNA itself: what variants are present and how genomes relate. When the question is instead which genes are active, the corresponding workflow is expression-based. Our guide to RNA sequencing analysis covers measuring gene expression, single-cell RNA-seq analysis resolves it to individual cells, and transcriptomics analysis covers the broader expression methods and their interpretation. Choosing DNA versus RNA sequencing is the first design decision, and it follows directly from whether your question is about sequence or about activity.

Variant Calling Done Properly: GATK Best Practices

A trustworthy germline call set follows a sequence of steps whose order is not arbitrary. After BWA-MEM alignment you mark duplicates so that the same PCR fragment is not counted as independent evidence, then run base quality score recalibration (BQSR) against a known-sites resource to correct the systematic miscalibration of the sequencer's per-base quality estimates. Calling itself uses GATK HaplotypeCaller, which performs a local de novo assembly of candidate haplotypes in each active region rather than judging one position at a time; this is what gives it a real advantage on insertions and deletions, where position-based callers stumble. For more than one sample, call each into a per-sample GVCF and then joint-genotype the cohort together with GenotypeGVCFs, which scales to large studies and sharpens calls at rare sites by sharing information across samples.

Filtering is not optional, because a raw VCF is dominated by artifacts. For large cohorts, variant quality score recalibration (VQSR) fits a Gaussian mixture model to annotations such as QD, MQ, FS, and ReadPosRankSum and assigns each variant a tranche-based probability of being real; for a single sample or a small project where VQSR cannot be trained, documented hard filters on those same annotations are the accepted substitute. Coverage governs what is even detectable: roughly 30x is standard for germline whole genome sequencing, whereas somatic and low-frequency detection needs far greater depth.

bwa mem -R '@RG\tID:s1\tSM:s1\tPL:ILLUMINA' ref.fa r1.fq r2.fq | samtools sort -o s1.bam
gatk MarkDuplicates -I s1.bam -O dedup.bam -M metrics.txt
gatk BaseRecalibrator -I dedup.bam -R ref.fa --known-sites known.vcf -O recal.table
gatk HaplotypeCaller -I dedup.bam -R ref.fa -ERC GVCF -O s1.g.vcf.gz

Beyond SNVs: Structural Variants, Copy Number, and Somatic Calling

Single-nucleotide variants and small indels are only one layer of genome variation. Structural variants, the deletions, duplications, inversions, and translocations that often carry the largest phenotypic effect, are detected from split reads, discordant read pairs, and local assembly by callers such as Manta, DELLY, or GRIDSS, while copy-number changes are inferred from read depth with tools like CNVkit. Cancer genomics adds a further distinction: separating somatic mutations acquired by the tumor from the germline variants present in every cell requires a paired tumor-normal design analysed with a somatic caller such as Mutect2, supported by a panel of normals to filter recurrent artifacts and by attention to tumor purity and the variant allele fraction that signals clonal versus subclonal mutations. Where short reads cannot resolve large rearrangements or repetitive and low-complexity regions, long-read platforms (PacBio HiFi, Oxford Nanopore) recover structural variants and enable haplotype phasing.

Annotation, Interpretation, and Bacterial Typing

A called variant is meaningless until it is placed in context. VEP or SnpEff predict the functional consequence, population databases such as gnomAD filter out common variants unlikely to drive a rare phenotype, and clinical interpretation weighs the ACMG criteria together with in-silico scores such as CADD or REVEL and curated evidence from ClinVar. Bacterial work reframes the whole problem as comparison rather than reference deviation: assess assembly quality with metrics like N50 and completeness checks (BUSCO or CheckM), type strains by multilocus sequence typing, and detect resistance and virulence genes with AMRFinderPlus, ResFinder, or the CARD database. For phylogenetics, build a core-genome alignment and then remove recombinant regions with Gubbins or ClonalFrameML before tree-building, because recombination violates the assumptions of standard phylogenetic models and distorts both branch lengths and topology; only then fit a maximum-likelihood tree with RAxML or IQ-TREE, optionally dated for transmission inference.

What a Genome Analysis Project Delivers

A complete genome sequencing analysis ships as a package: a quality-control report, the reproducible pipeline with every tool and reference build pinned, the aligned data, a filtered and annotated variant set (or an assembly and phylogenetic tree for bacterial work), downstream tables for your specific question, and methods text ready for your manuscript. When the result needs to become a figure, an annotated tree, a variant landscape, or a coverage plot, our research figure service produces final, publication-compliant graphics.

When you are ready, get a free quote with a short description of your organism, sample number, and design, or explore the full list of research services to combine genome analysis with statistics, visualization, or manuscript support.

Free sequence tools

Work with your sequences directly: reverse complement a strand, check GC content and skew, translate DNA to protein across all six frames, and compare two sequences with a dot plot. For a complete analysis pipeline, see the bioinformatics analysis service.