Droplet Biosciences Accelerates Residual Cancer Detection With NVIDIA Parabricks

By collecting lymphatic fluid 24 hours after a patient’s tumor is removed, Droplet Biosciences has changed the timeline and process of how residual cancer is detected after surgery. Unlike blood tests, which are widely used for minimal residual disease (MRD) monitoring, lymphatic fluid not only allows for earlier collection but also increases sensitivity.

In particular, the team at Droplet Biosciences has found that post-surgical lymphatic fluid has 130x more tumor-derived DNA molecules than time-matched blood plasma. The significant increase in tumor-derived DNA molecules can be attributed to being local to the tumor site rather than being diluted in liters of blood.

“By sampling fluid that’s closer to where the tumor was growing, we have an increased concentration of tumor-associated molecules, and that’s translating to increased sensitivity to identify patients with residual disease,” explains Wendy Winckler, PhD, Chief Scientific Officer at Droplet Biosciences. “That’s important at a time point that allows clinicians to tailor patients' therapeutic regimens, for example, to give additional therapy for those patients who still have residual disease.”

As tumors grow and spread, they will eventually test positive and show in MRD monitoring in blood. However, these first tests typically aren’t done until four to six weeks post-surgery when enough time has passed for residual disease to become detectable in blood. Droplet Biosciences is accelerating this timeline to just 24 hours post-surgery—resulting in faster turnaround time that ultimately helps influence and improve care.

This is particularly relevant to rare residual cancer cells. As Winckler explains, "finding them is like looking for a needle in a haystack. If your sample contains many more DNA fragments—and a higher concentration of tumor-derived DNA—you’re simply more likely to capture that ‘needle’ when you test. That’s translating to about double the sensitivity of time-matched blood.”

Droplet Biosciences’ first clinical test is a Clinical Laboratory Improvements Amendments (CLIA) validated clinical test for HPV negative head and neck cancer. Additionally, they also have late-stage programs for lung and bladder cancer that are currently going through proof of concept. Collecting lymphatic fluid is applicable across solid tumors and can apply to a wide variety of procedures and cancers in the future.

Providing faster turnaround times also directly impacts the patient experience. Instead of having to wait weeks and come back to the hospital after surgery, patients are able to take a residual cancer diagnostic test while still at the hospital and avoid waiting to come back for traditional blood tests. It also helps provide better peace of mind that clinical teams are able to detect residual molecules after surgery with a high degree of confidence because of the work from Droplet Biosciences.

To accelerate analysis, Droplet Biosciences implemented NVIDIA Parabricks, a software suite for genomic analysis that provides GPU-accelerated versions of open-source tools.

“We are using Parabricks to speed up our genomic analysis and shorten turnaround time from 10 days to less than five days,” says Zhuosheng Gu, Senior Director of Informatics, R&D at Droplet Biosciences. “We’re in the process of implementing more GPU-accelerated tools and should be able to shorten our analysis to less than two days with everything implemented in the next six months.”

In order to detect rare variants, lymphatic fluid samples require significant NGS coverage—resulting in 5000x coverage after deduplication with raw coverage typically over 10000x. Without GPU acceleration, sequencing to this coverage depth can result in significant computational challenges.

“It took a lot of computational power to process these samples in the CPU environment," explains Gu. “Currently, we’re using the Parabricks FQ2BAM (fastq2bam) step to speed up alignment. On CPU, one sample could take up to 36 hours. With the help of GPU acceleration, we were able to shorten that particular step to less than three hours on average.”

A member of NVIDIA Inception, an AI startup accelerator-program, and an NVIDIA AI Enterprise customer, Droplet Biosciences has received technical support and also worked directly with the Parabricks team on product feedback as well as early access to containers to try and test new features in their environment.

"It's also been a help to the team in the clinical testing lab because our timeline now more comfortably allows for rework. With a 10-day test process and a 10-day target turnaround time, things had to go right every single time. Now with a faster pipeline and a quicker read on which samples passed, we're able to potentially go back and repeat a step in the lab and still have time to process it going through the much faster pipeline. It's been really game changing for us."

Looking ahead to what's next for the team, they are focusing on speed without compromising accuracy. Droplet Biosciences currently has a base error model that requires significant processing to quantify background noise at the position of each mutation. "Previously, just processing that step alone took hours," explains Gu. "After switching to GPU, we can build the base error model for new tumor mutations within 30 minutes for each sample."

As future possibilities continue to grow, focusing on efficiency remains important to the core of Droplet Biosciences’ work. “The ability to speed up our turnaround time this significantly—going from five days to potentially one to two—has had really important impacts on being able to deliver on our value proposition of tailoring personalized care at the adjuvant setting,” explains Winckler.

“The ability to speed up our turnaround time this significantly—going from five days to potentially one to two—has had really important impacts on being able to deliver on our value proposition of tailoring personalized care at the adjuvant setting.”

Wendy Winckler, PhD
Chief Scientific Officer
Droplet Biosciences