a16z Podcast: Revisiting the Gene

Episode Summary

Executive Summary: The panel explores how genomics is moving from cheap sequencing to costly interpretation and dynamic diagnostics. Carlos Oraya argues that most value lies in contextual variant interpretation for inherited disease risk, while Gabe Ott explains Freenome’s approach to reading dynamic DNA from blood as an AI-driven, longitudinal signal for disease detection, especially cancer. The discussion also highlights clinical misinterpretation risks and unresolved reimbursement challenges.

Main Topics: From sequencing to application layers (Priority: 5/5): The conversation frames genomics as having moved beyond generating DNA data into building clinical applications that make the data actionable. Variant interpretation in inherited disease (Priority: 5/5): Carlos Oraya explains Jungla’s focus on contextualizing variants, distinguishing disease-risk variants from causal ones, and building models to support interpretation teams. Dynamic DNA and liquid biopsy diagnostics (Priority: 5/5): Gabe Ott describes Freenome’s thesis that blood-derived DNA reflects ongoing biological processes, enabling snapshots of molecular health rather than static genetic predisposition. AI/ML for diagnostic signal extraction (Priority: 4/5): Because dynamic DNA signals are noisy and convoluted, machine learning is presented as essential for identifying disease-specific patterns and improving diagnostic performance over time. Clinical risk, misinterpretation, and patient harm (Priority: 4/5): The panel discusses how unclear or poorly communicated genomic results can lead to inappropriate clinical decisions, underscored by an Oregon case involving unnecessary surgery. Reimbursement and adoption barriers (Priority: 4/5): Even if diagnostics are scientifically promising, reimbursement remains difficult because payers need clear return-on-investment evidence before covering tests.

Key Arguments: Sequencing has become cheap and fast, but interpretation remains expensive and clinically difficult; the bottleneck has shifted from data generation to meaning-making. For inherited conditions, genomic interpretation must be contextualized using family history, other tests, and condition-specific relevance rather than reading variants in isolation. Most clinical interpretation today is concentrated in the genetic test provider backend, where specialized teams can classify variants more safely than general physicians. Only a tiny fraction of possible disease-gene mutations have clinical interpretations, showing how incomplete current genomic knowledge remains. Dynamic DNA in blood can reveal real-time biological change, unlike static inherited DNA, making it more suitable for monitoring disease activity. AI-based diagnostics can improve after launch by learning from real-world test results, potentially reversing the usual decline in performance after clinical rollout. False positives and false negatives are major limitations of traditional diagnostics; large-scale, AI-assisted feedback loops may help address them. Adoption depends not only on scientific validity but on reimbursement models that demonstrate cost savings or value to payers.

Data Points: Original Human Genome Project cost: $3 billion - The first human genome project required about this amount to generate a single human genome. Original Human Genome Project duration: 13 years - The first human genome project took this long to complete one genome. Current genome sequencing cost: about $1,000 - Today, the same sequencing task can be done for roughly this amount in a couple of days. Current genome sequencing cost trend: hundreds of dollars per genome - Oraya notes the cost is now falling into the hundreds for some sequencing workflows. New variants per genome: on the order of 3 million - Each completed human genome sequence identifies roughly this many new variants. Novel variants in disease-associated genes: roughly 100 variants - These are the variants that require interpretation under current clinical practice. Interpretation cost per variant: $50 to $100 - Current clinical interpretation of each novel disease-associated variant costs this much. Interpretation vs acquisition cost gap: 100 to 1,000-fold - Interpretation can cost far more than data acquisition, according to Oraya. Clinical interpretations coverage: roughly 0.6% - Only this fraction of possible mutations in known disease-associated genes have clinical interpretations. 23andMe DNA coverage: less than 1% - Ott uses this to emphasize that consumer genotyping captures only a tiny portion of DNA information. White blood cell turnover: literally turning over every day - Used as an example of dynamic biology despite identical DNA. DNA floating in blood turnover: every 20 minutes - Ott describes blood-derived DNA as a fast-changing signal of bodily processes. Mammography false positive rate: 50% - Ott cites this as an example of how current screening can be highly error-prone. Largest clinical trial announced (not yet performed): 120,000 people - Ott uses this to contrast traditional trial size with the scale of real-world data collection. U.S. colorectal cancer screening gap: 35 million people - Ott says this many people should have been screened in the U.S. last year but were not. Typical reimbursement rate for diagnostic tests: about 20% - Ott says only about one-fifth of tests sold are fully reimbursed. Tests not properly paid for: 80% - The remainder are often not reimbursed adequately, creating adoption friction.

Pivotal Quotes: "Reading the DNA isn't the same thing as understanding it." — Jorge Condé: He introduces the core shift from sequencing capacity to interpretation and application. "DNA is not static. DNA is actually incredibly dynamic and it changes in all sorts of ways." — Gabe Ott: He explains Freenome’s premise that blood-based dynamic DNA can reveal ongoing disease states. "For the first time, we have an opportunity to make the direction of the accuracy of a test after we launch it go up as opposed to go down." — Gabe Ott: He argues AI-enabled diagnostics can improve with post-launch feedback, unlike traditional tests.

Implications: Genomics is shifting from data generation to clinically actionable interpretation and real-time monitoring. The winners will need strong validation, AI-assisted learning, careful physician communication, and reimbursement strategies that prove economic value.