DRAGEN MRD Pipeline
For command line examples and parameter details, see DRAGEN recipes: DNA Somatic Tumor-Normal-MRD
The DRAGEN MRD (Minimal Residual Disease) pipeline is a tumor-informed whole genome sequencing (WGS) research workflow designed to detect tumor-derived circulating tumor DNA (ctDNA) in plasma samples. The workflow uses genomic sites and alleles previously identified as somatic variants in a matched tumor sample. The variants are collectively referred to as the tumor fingerprint. Specialized algorithms identify rare tumor-derived molecules and support detection of ctDNA at very low tumor fractions. The MRD Detect component analyzes plasma sequencing data using the tumor fingerprint and generates quality control (QC) metrics to help assess sequencing run and data quality. The resulting MRD measurements can be used to assess changes in ctDNA levels over time.
Baseline Test: Tumor Fingerprint
The tumor fingerprint profile is typically established from a biopsy of primary tumor tissue paired with a matched normal sample. The DRAGEN small variant caller identifies high-confidence somatic variants unique to the tumor sample from the matched sample pair. These variants collectively constitute the tumor fingerprint. Illumina recommends sequencing DNA from Formalin-Fixed Paraffin-Embedded (FFPE) or fresh frozen tumor tissue to achieve a target raw coverage of 100X, and DNA from buffy coat or plasma-depleted whole blood (BC) as the matched normal sample to achieve a target raw coverage of 50X.
Follow-Up Test: Plasma Follow-Up
Follow-up plasma samples are collected over time for analysis of tumor-derived DNA. The tumor fingerprint is used to identify the variant sites analyzed by the MRD workflow. MRD Detect generates sample QC metrics related to the coverage of informative sites in the tumor fingerprint profile and provides noise estimates for a given plasma sample. Illumina recommends sequencing plasma samples to achieve a target raw coverage of 50X.
Pipeline
The DRAGEN MRD pipeline does not include a pre-built workflow script. Instead, it defines the required computational steps. Data management, sample tracking, and workflow scripts are left to the user.
BCL demultiplexing must be completed before the DRAGEN MRD pipeline can be run to ensure that sample-specific FASTQ files are available for pipeline input.
The diagram below illustrates the two primary workflows: Fingerprint Workflow and MRD Workflow. Detailed steps are outlined in the accompanying table.

Step 0: FASTQ generation
Convert BCL files to FASTQ format for the FFPE, BC, and plasma samples.
Step 1: Read alignment and targeted germline variant calling
Map and align reads from the FFPE/BC/Plasma trio samples to generate BAM/CRAM files and call germline variants at ~37K single nucleotide polymorphism (SNP) sites with high population allele frequencies (typically close to 50% VAF) to generate germline VCF files. The selected sites for germline variant calling have high variability and are informative for sample matching and cross-contamination QC (i.e., contamination from a different human) algorithms. This step also provides BC sample cross-contamination QC results.
Step 2: Fingerprint generation + QC
(A) Run the DRAGEN Tumor/Normal somatic small variant caller on the matched FFPE-BC sample pair to generate a fingerprint VCF. This step also provides FFPE normal-aware sample cross-contamination QC results. (B) Run FFPE/BC sample matching QC using the DRAGEN CheckFingerprint module, which compares the FFPE and BC germline VCFs generated in Step 1 to determine whether they are from the same individual using a statistical test based on matching alleles. A positive LOD (Logarithm of the Odds) score indicates that the two samples are from the same individual.
Step 3: MRD detection
Run the MRD Detect module on the plasma sample using the tumor fingerprint VCF to analyze tumor fingerprint sites and generate MRD results.
Step 4: Plasma QC
(A) Run plasma/BC sample matching QC using the DRAGEN CheckFingerprint module, which compares the plasma and BC germline VCFs generated in Step 1 to determine whether they are from the same individual using a statistical test based on matching alleles. A positive LOD (Logarithm of the Odds) score indicates that the two samples are from the same individual. (B) Run the MRD Detect module on the plasma sample using ~37K SNP sites with high population allele frequencies (typically close to 50% VAF) after excluding the individual's BC germline variants to identify very low levels of contamination from another human sample.
Step 0: FASTQ generation
Convert BCL files to FASTQ format for the FFPE, BC, and plasma samples. This step is optional if FASTQ data already exists. If starting from BCL files, this step must be completed before running the pipeline to ensure that sample-specific FASTQ files are available as input.
For command line details, see the MRD recipe: Step 0
Step 1: Read alignment and targeted variant calling
Map and align reads from the FFPE/BC/Plasma trio samples to generate BAM/CRAM files and call germline variants at ~37K single nucleotide polymorphism (SNP) sites with high population allele frequencies (typically close to 50% VAF) to generate germline VCF files. These germline VCF files are used for sample matching and plasma contamination detection. Illumina recommends the graph/pangenome reference for germline variant calling, but MRD analyses are performed on the linear reference for consistency.
For FFPE samples, use Step 1A with the --Aligner.hard-clips=7 parameter. For BC and plasma samples, use Step 1B.
**For command line details, see the MRD recipe: Step 1A (FFPE) and Step 1B (BC/Plasma)**
Step 2: Fingerprint generation + QC
Step 2A: Fingerprint generation and FFPE normal-aware contamination QC
Run the DRAGEN Tumor/Normal somatic small variant caller on the matched FFPE-BC sample pair to generate a fingerprint VCF. The setting --mrd-fingerprint=true activates additional strict filters, including more aggressive read position filtering. This helps improve specificity of variant calls in the tumor fingerprint. This step also provides FFPE normal-aware sample cross-contamination QC results.
For command line details, see the MRD recipe: Step 2A
Step 2B: FFPE/BC sample matching QC
Run FFPE/BC sample matching QC using the DRAGEN CheckFingerprint module, which compares the FFPE and BC germline VCFs generated in Step 1 to determine whether they are from the same individual using a statistical test based on matching alleles. A positive LOD (Logarithm of the Odds) score indicates that the two samples are from the same individual.
For command line details, see the MRD recipe: Step 2B
For more details on sample matching (also referred to as DRAGEN CheckFingerprint, but not to be confused with the tumor fingerprint), refer to: Check Sample Identity with CheckFingerprint.
Step 3: MRD detection
Run the MRD Detect module on the plasma sample using the tumor fingerprint VCF to analyze tumor fingerprint sites and generate MRD results. In this step, lower-quality reads are excluded from the analysis, such as reads with low Phred scores or reads supported by only one read direction. Variant signals observed at the individual's fingerprint sites are counted and compared to a statistical noise model based on dynamically generated noise sites to produce a statistical log-likelihood score. When the signal at fingerprint sites exceeds the level expected from sequencing noise, the workflow generates a detection result. Higher scores indicate stronger evidence of tumor-derived DNA in the analyzed sample.
For command line details, see the MRD recipe: Step 3
Interpreting eVAF
In addition to the log-likelihood score, the MRD Detect module reports an estimated Variant Allele Frequency (eVAF) as a second key quantitative metric. eVAF reflects the allele-level frequency of detected variants in the plasma sample and can be used to assess ctDNA signal and associated changes over time.
eVAF vs. tumor purity/content
eVAF is an allele-level measurement and should not be confused with tumor purity, a traditional pathology term (also referred to as tumor content, tumor fraction, or cellularity) that describes a cell-level measurement: the percentage of tumor cells among all nucleated cells in the selected tissue area.
In the context of molecular testing, tumor purity/content/fraction can also refer to the proportion of DNA molecules in the sequenced sample that originate from tumor cells. These two quantities are related but not equivalent: because tumor cells are often aneuploid, tumor DNA fraction is typically greater than tumor cellularity. For example, a sample with 30% tumor cellularity may yield 40% tumor DNA molecules due to the elevated copy number content per tumor cell.
The mathematical relationship between VAF, tumor purity, and copy number is (Birkbak et al., 2014):
where is tumor cellularity (i.e., tumor purity in the pathology sense), is the multiplicity (number of copies of the variant allele per tumor cell), is the copy number in tumor cells, and is the copy number in normal cells.
Practical example: heterozygous mutation, diploid tumor
For a heterozygous somatic mutation with no major copy number changes, substituting , , and simplifies the formula to:
Under these conditions, the observed VAF is half the tumor purity. For example, a 50% tumor purity sample produces an eVAF of 25%. In the MRD context, this relationship allows eVAF values observed at fingerprint sites to be used as a direct estimate of ctDNA fraction in the absence of copy number alterations.
Step 4: Plasma QC
Step 4A: Plasma/BC sample matching QC
Run plasma/BC sample matching QC using the DRAGEN CheckFingerprint module, which compares the plasma and BC germline VCFs generated in Step 1 to determine whether they are from the same individual using a statistical test based on matching alleles. A positive LOD (Logarithm of the Odds) score indicates that the two samples are from the same individual.
For command line details, see the MRD recipe: Step 4A
Step 4B: Plasma contamination QC
Run the MRD Detect module on the plasma sample using ~37K SNP sites with high population allele frequencies (typically close to 50% VAF) after excluding the individual's BC germline variants to identify very low levels of contamination from another human sample. Specifically, the algorithm analyzes pileups at the specified loci with high population allele frequencies after excluding germline variant sites known to be present in the individual, and then doubles the estimated VAF (eVAF) at these loci to approximate the fractional foreign contamination.
This module does not inherently adjust for VAF distortions that can occur in copy number variation (CNV)-rich somatic samples. Therefore, Illumina recommends including a safety margin when considering reported plasma contamination levels.
For command line details, see the MRD recipe: Step 4B
Interpreting eVAF in plasma contamination QC
In the plasma contamination QC workflow, the MRD Detect module measures signal at loci with population allele frequencies close to 50%. At these sites, a foreign contaminant is expected to contribute alleles at approximately half the rate of the foreign DNA fraction—analogous to the heterozygous diploid case described above. The inferred foreign DNA fraction is therefore approximately:
For example, an eVAF of 0.001 (0.1%) at the contamination SNP sites implies approximately 0.2% foreign DNA in the plasma sample.
Last updated
Was this helpful?

