How to Read the GATK Best Practices
Why Is This Concept Necessary
Identifying variants in sequencing data does not conclude with a single command. The final variant call is influenced by the alignment of raw reads to positions in the reference genome and their associated qualities, the consistency of observed bases at those positions, and how multiple samples are compared.
GATK Best Practices are frequently used as a reference point for understanding such variant analysis workflows. However, the term "Best Practices" does not imply a fixed recipe that applies uniformly across all datasets and GATK releases.
This document serves as a guide to navigating the boundary between concepts and evidence. Step-by-step tutorials continue in the GATK4 Best Practices Hands-on Series.
Core Concept — A Pipeline Is Not a List of Commands, but a Process for Transforming Evidence
The conceptual workflow is as follows:
sequencing reads
↓
alignment to a reference genome
↓
quality control and preprocessing
↓
variant evidence aggregation
↓
variant calling and filtering
↓
VCF + quality / evidence annotationsEach step constitutes both a process of transforming one file into the next and a reinterpretation of the observed evidence. For example, the alignment of reads to a BAM file does not imply that all mismatches represent biological variants; sequencing errors, mapping ambiguity, duplicates, and low coverage must all be jointly considered.
Relationship Between Input and Output
FASTQ
FASTQ files contain read sequences and base quality scores. At this stage, since there are no reference coordinates, it is not possible to directly determine "which gene has which variant."
BAM/CRAM
Aligning reads to the reference genome allows us to verify where each read is mapped and its alignment quality. Factors such as the reference build, aligner, read group, and duplicate handling influence the interpretation in subsequent steps.
VCF
Variant callers aggregate allelic evidence at specific positions from aligned reads to record variant candidates and their quality information. The VCF format represents a summary of results rather than the final truth table, as all raw data and decision processes are not preserved. It should be interpreted alongside information regarding the reference genome, caller parameters, filtering criteria, depth, and allele balance.
Are germline and somatic analyses performed using the same pipeline?
No. Germline variant calling addresses the question of identifying an individual’s inherited genetic variants, whereas somatic variant calling aims to detect differences between tumor and normal samples or identify intratumoral heterogeneity. Differences may arise in sample composition, expected allele fractions, error models, filtering criteria, and interpretation standards.
Consequently, simply stating “GATK pipeline” is insufficient. The following details must be explicitly provided:
- Whether the analysis targets germline or somatic variants
- Whether the approach is single-sample or cohort-based
- Whether whole-genome sequencing (WGS), whole-exome sequencing (WES), or a targeted panel was used
- The reference genome build employed
- The GATK release version and upstream/downstream tools used
For germline cohort analysis, the conceptual workflow should be delineated as analysis-ready BAM → per-sample gVCF → gVCF aggregation → joint genotyping → filtering. In contrast, somatic short variant workflows constitute a distinct analytical question that takes tumor samples and optionally matched normal samples as inputs.
Small Example — Interpreting the Same Mismatch Differently
Suppose that at a given position, 15 out of 20 reads show a nucleotide different from the reference. This result alone is insufficient to definitively call a variant.
| Check Item | Question |
|---|---|
| Read depth | Is sufficient read coverage observed? |
| Base/mapping quality | Are the base and mapping qualities for this position reliable? |
| Strand balance | Is there bias toward reads from only one strand? |
| Allele fraction | Does the allele fraction align with germline or somatic expectations? |
| Nearby alignment | Are there indels, repetitive regions, or mapping ambiguity nearby? |
Only after completing these checks can an observed difference be interpreted as variant evidence. Although pipelines automate this process, automation does not eliminate the need to define quality criteria or assume responsibility for quality control.
Why Versioning Matters
The steps and recommendations in the GATK Best Practices can vary depending on the data type, GATK release, reference genome, variant caller, and parameters. Mixing step names from previous articles with those of current official workflows within a single document may inadvertently present historical educational pipelines as current, reproducible methodologies.
A reproducible record must include the following:
- The exact GATK release
- The exact version or commit hash of the executed workflow/WDL
- Workflow or tool versions
- Reference genome build and known-sites resources
- Input data types and sample design
- Filtering and quality control criteria
- VCF post-calling annotation and interpretation resources
This concept article does not provide specific GATK execution commands. Dataset-specific versions and conditions can differ, so it focuses on the data flow and decision points.
Points of Confusion
Are Best Practices a permanent standard recipe?
No. Best Practices serve as benchmarks for validated workflows tailored to specific analytical questions and tool versions. Before applying them currently, you must reconsult the official documentation and release-specific guidance.
Do variants recorded in a VCF represent clinically significant mutations?
No. A VCF (Variant Call Format) is an analytical output that contains variant calls and associated evidence. Clinical significance requires a separate interpretation step involving annotation, population data, phenotype information, and established guidelines.
Do errors disappear once preprocessing is complete?
No. Preprocessing is a process designed to reduce specific errors and biases; it does not guarantee the elimination of all errors. Limitations related to coverage, mapping, sample quality, and assay design may persist.
Current Evidence and Limitations
- GATK can be described as an ecosystem of tools and workflows that constitute NGS variant analysis pipelines.
- Alignment, preprocessing, variant calling, and filtering are interconnected stages.
- No assumption is made that specific historical workflows align with current official workflows.
- Indel realignment is not presented as a mandatory GATK4 preprocessing step in this context.
- This article is not a command tutorial; it is a conceptual guide to inputs, evidence, outputs, and version dependencies.
Connection Concept / Story
- Concept: FASTQ, BAM/CRAM, VCF, reference genome, variant calling
- Story candidate: The analytical process through which a single read mismatch is narrowed down as variant evidence