Healthcare and Life Sciences
DELFI Diagnostics Accelerates Cancer Screening and Detection with Parabricks
DELFI Diagnostics
Objective
With lung cancer as the leading cause of cancer death in the United States, DELFI Diagnostics, an early-stage cancer diagnostics company, is working to close the gap in lung cancer screening and improve survival rates. To support its mission of making cancer detection faster and more accessible, DELFI has evaluated incorporating NVIDIA Parabricks to accelerate its genomic analysis pipelines—and demonstrated that incorporating this technology can achieve 6.6x faster compute time and an 8.1x reduction in compute cost.
Customer
DELFI Diagnostics
Partner
Accelerated Computing Tools & Techniques
Products
Key Takeaways
- When evaluating NVIDIA Parabricks for sequence alignment with their FirstLook Lung product, DELFI achieved 5.4x faster compute time and a 4.1x cost reduction.
- When evaluating NVIDIA Parabricks for sequence alignment and mutation calling with their DELFI-TF product, DELFI achieved 6.6x faster compute time and an 8.1 cost reduction.
Leveraging Liquid Biopsy for Cancer Detection
DELFI Diagnostics is an early-stage diagnostics company using liquid biopsy to make cancer detection more accessible. Founded in 2019 and rooted in Johns Hopkins research, DELFI uses machine learning and AI to analyze cell-free DNA (cfDNA) fragments in the blood to offer a minimally invasive, low-cost approach to cancer screening.
Unlike traditional tissue biopsies, which require a surgeon to remove tumor tissue, liquid biopsy requires only a standard blood draw. Because blood circulates throughout the body, these assays can capture DNA shed from tumors across multiple sites—rather than sampling a single location. Liquid biopsy is well suited to screening and repeat testing over time.
“A few things set us apart, but the most important is the foundation of our technology. Most liquid biopsy approaches look for specific mutations or biomarkers, which means they’re searching for known signals in a relatively small portion of the genome,” says Jen Bruursema, Senior Director of Marketing and PR at DELFI Diagnostics. “Our approach is different. We analyze genome-wide patterns in cell-free DNA, which gives us a much broader view of what the biology is telling us.”
DELFI’s lead product, FirstLook™ Lung, is an AI-powered lung cancer screening test that uses low-pass whole-genome sequencing (WGS). FirstLook Lung is designed to reach a large subset of the population eligible for lung cancer screening who have not participated in traditional low-dose CT screening. The test helps address this gap in lung cancer screening by providing health systems and clinicians with a new, low-cost, non-invasive way to engage this patient group.
FirstLook Lung received a Clinical Laboratory Permit and Test Approval letter from the New York State Department of Health (NYSDOH) in May 2026. NYSDOH approval completes authorization for DELFI to serve health systems across all 50 U.S. states.
DELFI Diagnostics
“For FirstLook Lung specifically, we’re the only blood-based, commercially available lab-developed test for lung cancer screening right now,” Bruursema explains. “That matters because our mission is accessibility. A test only helps patients if they can actually get it.”
Additionally, DELFI offers a research service for pharmaceutical companies, DELFI-TF (Tumor Fraction), a tumor-agnostic method to monitor treatment response in oncology trials and guide decisions in drug development. Using less than 1 mL of plasma, DELFI-TF is built on machine learning trained on a large set of longitudinally treated samples and operates at a far lower cost than traditional methods.
“Turnaround time becomes even more critical because it allows more personalized cancer treatment to be done in near real time.”
Kevin Jacobs
Sr. Director, Software Engineering, DELFI Diagnostics
DELFI Diagnostics
When Speed Matters Most
Reducing turnaround time is critically important for cancer patients at every stage, from informing treatment decisions and monitoring how tumors respond to detecting recurrence as early as possible. DELFI-TF's current turnaround time is 10–20 business days. As DELFI looks to transition therapeutic monitoring from a research service to a commercial, clinical product, reducing this turnaround time will be a key driver for clinical utility.
While shipping and sequencing have fixed time constraints, the bioinformatics portion of the pipeline can be optimized, making it an actionable area for improvement. As DELFI-TF scales to clinical volumes, processing more samples will require more compute, and GPU acceleration positions the pipeline to handle that growth without causing turnaround time to slip. It also opens the door to deeper sequencing coverage down the line. Although DELFI-TF currently runs at 4x depth, a test study with Ultima sequencing generated data at approximately 150x depth (AACR 2026), and GPU acceleration makes that level of depth computationally practical to consider for future versions of the assay.
A critical component of disease monitoring is the identification of somatic variants, which are acquired, rather than inherited, DNA mutations closely associated with cancer. Because somatic variant calling is computationally expensive, it has historically been cost-prohibitive to run at scale. However, GPU acceleration can directly reduce compute time and cost, bringing total turnaround time from 10–20 business days down to 5–7 days.
“The more we can shorten overall turnaround time, the more likely we are able to deliver the results at an impactful time.”
Nidhi Narang
Vice President, Software and Data Engineering, DELFI Diagnostics
Accelerating Analysis With NVIDIA Parabricks
Prior to evaluating NVIDIA Parabricks, DELFI used open-source tools. Although these tools are industry standard as the foundation of most genomic analysis pipelines, processing high sequencing volumes on CPU infrastructure can introduce bottlenecks in compute time. NVIDIA Parabricks, a genomics software suite for accelerated secondary analysis, provides GPU-accelerated versions of many open-source tools.
Parabricks’ FQ2BAM is a GPU-accelerated tool that replaces the CPU-based alignment and preprocessing steps—including BWA-MEM, GATK SAM sorting, duplicate marking, and optional BQSR. Slotting GPU versions of these tools directly into existing pipelines enables DELFI to achieve faster results without rewriting or redesigning workflows, since inputs and outputs remain unchanged.
For somatic variant calling, Parabricks provides GPU-accelerated tools like MutectCaller (the Parabricks version of open-source Mutect2) and Google’s DeepSomatic. Previously, somatic variant calling was cost-prohibitive on CPU infrastructure, but GPU-accelerated somatic variant calling enables genome-wide variant detection, giving clinicians a clearer picture of which cancer clones are recurring or responding to treatment.
For FirstLook Lung, running Parabricks on a low-pass WGS pipeline yielded 5.4x faster compute time and a 4.1x reduction in cost for sequence alignment. “Utilizing GPU technology would enable DELFI to deliver low-cost, high-volume testing with rapid turnaround times,” explains Bruursema. “This significant reduction in time and cost has the potential to expand patient access and equip doctors with the fast insights needed to initiate treatment sooner.”
For DELFI-TF, the higher coverage depths result in even stronger improvements—including 6.6x speedup and an 8.1x cost reduction for sequence alignment and somatic mutation calling. Bruursema elaborates, “When DELFI-TF utilizes deep WGS, this reduction in processing time results in a direct, substantial decrease in total computation costs.”
“Utilizing GPU technology enables DELFI to deliver low-cost, high-volume testing with rapid turnaround times. This significant reduction in time and cost has the potential to expand patient access and equip doctors with the fast insights needed to initiate treatment sooner.”
Jen Bruursema
Senior Director of Marketing and PR, DELFI Diagnostics
Looking Ahead
DELFI’s collaboration with NVIDIA extends beyond software integration. “As a member of the NVIDIA Inception program, the DELFI Diagnostics team worked directly with the Parabricks team to integrate the Parabricks framework into DELFI workflows,” says Bruursema. “The technical support received from the NVIDIA team has been instrumental in the successful integration of NVIDIA Parabricks within DELFI products.”
Specifically for DELFI-TF, integrating Parabricks is a key step in planning the transition from a research service to a commercial clinical product. For clinicians, faster turnaround times enable more timely treatment decisions because speed directly impacts patient outcomes. It also unlocks previously time- and cost-prohibitive capabilities, such as somatic variant calling, creating new possibilities for disease monitoring.
“We do not view the infrastructure and partnerships enabling our computational efforts as mere administrative costs; they are fundamental to our scientific process,” says Amoolya Singh, CTO, DELFI Diagnostics. “Our methodology relies entirely on the capacity to perform rapid, large-scale analysis of genome-wide data.”
As WGS coverage depths and sample volumes grow, accelerated computing will only continue to improve speed and reduce cost—supporting DELFI’s mission of making cancer detection accessible to more patients.
Learn more about NVIDIA solutions for genomics.