BillionToOne: from half a lab bench to cancer's holy grail

Executive overview

Most genetic tests amplify DNA from blood samples — but amplification adds noise that drowns out rare signals from a fetus or tumour. BillionToOne solved this by adding synthetic DNA to samples before amplification, using the known quantity to measure and subtract distortion. The result is a test sensitive enough to find one altered base pair among three billion.

Starting with prenatal testing, they reached 20% US market share and over 600,000 tests per year. The same core technology now targets cancer — first detecting residual disease after surgery, then eventually screening healthy patients for early-stage tumours.

The core insight: convert a biology problem into a maths problem by engineering a known reference into the sample before you touch it.

The noise problem and how BillionToOne solved it

• Fetal and tumour DNA in blood is extremely dilute — a few molecules among billions
• PCR amplification, required to make signals readable, also amplifies errors and noise
• BillionToOne adds proprietary synthetic DNA (quantitative counting templates) to each sample before amplification
• Because the synthetic quantity is known, the system can calculate exactly how much distortion was introduced
• Machine learning then strips that noise from the sequencing data
• What remains is a clean signal from the original sample

Building the first test: early constraints

• Two PhD founders — one at Stanford, one at Rice — combined biology and data science skills that rarely coexist in one person
• First lab space was a shared half-bench; early suppliers questioned whether the company could pay invoices
• First $300,000 raised took six months at $10,000 per cheque
• Working prenatal test developed and proven within six months of starting
• Two months after commercial launch, only one physician was sending one or two samples per week

Cracking sales

• Growth stalled because the team wasn't reaching physicians at scale
• Emergency decision: hire five additional sales reps in three weeks, train over a weekend, deploy on Monday
• Switched strategy to target patients directly — patients then pressured their own doctors to order the test
• One in five patients converted; inside sales reps spent 30–45 minutes per patient scripting exactly what to say to their doctor
• Traction from patient-led demand gave the company credibility to recruit experienced sales hires

Lab operations at scale

• Facility in Union City, California now processes over 600,000 tests per year; capacity projected to reach two million (roughly one in three US newborns)
• Samples arrive as raw blood; centrifuged to separate plasma, then liquid-handling robots extract only the cell-free DNA layer
• Samples from up to 1,000 patients are barcoded and pooled into a single sequencing run, then computationally de-multiplexed
• AI and computer vision reduced the sample intake step from 60 seconds per file (human) to near-instant
• Most results flow through automated analysis; complex cases go to a panel of genetic counsellors and laboratory directors

The three-step plan

• Step one (complete): prenatal carrier screening — least capital-intensive, fastest path to revenue and proof of technology
• Step two (in progress): minimal residual disease (MRD) testing for stage 3–4 cancer patients after treatment — detects microscopic tumour DNA that scans cannot see; launch expected within a year
• Step three: MRD testing for stage 1–2 patients after curative surgery, where ~20% harbour undetectable residual disease
• Step four (long-term): annual population screening to catch cancer before it reaches stage one — the "holy grail" the industry has chased for decades

Patient impact

• A patient in their forties with metastatic colorectal cancer had exhausted all treatment options and was entering hospice
• BillionToOne's liquid biopsy identified microsatellite instability across multiple tumour sites that a tissue biopsy had missed due to spatial sampling error
• The patient was matched to immunotherapy and responded dramatically; their oncologist now sends blood tests from all cancer patients

Team and research structure

• Hiring for interdisciplinary individuals, not interdisciplinary teams — one scientist who bridges chemistry and data science iterates faster than a handoff between specialists
• Small pods: one principal investigator plus two or three research associates, reporting directly to the founders
• Each pod owns end-to-end development of a single product, minimising bureaucracy
• Structure intentionally resembles multiple internal startups within the larger company