a16z Podcast | Revisiting the Gene

Cost and Capacity Shift in Genomics

• The cost to sequence a human genome dropped from approximately $3 billion over 13 years (Human Genome Project) to roughly $1,000 in a couple of days.
• Current sequencing can identify approximately 3 million new variants per genome, but only about 0.6% of possible mutations in disease-associated genes currently have clinical interpretations.
• While data acquisition is cheap, interpretation costs are 100 to 1,000 times higher than the cost of generating the raw data (estimated at $50–$100 per variant under current clinical practices).

Jungla's Approach to Data Interpretation

• Jungla focuses on contextualizing genetic data by linking variants to specific conditions, family history, and other tests to distinguish meaningful "trees" from the genomic "forest."
• The company builds computational models to predict molecular and cellular effects for the ~9.4% of variants currently of unknown significance.
• Diagnostic guidance provided to physicians includes clear statements when a variant's effect is unknown, alongside audit-tracked diagnostic metrics.

Freenome's Dynamic DNA Strategy

• Freenome analyzes dynamic DNA fragments circulating in blood (turning over every ~20 minutes) rather than static germline DNA (used by tests like 23andMe which sample <1% of DNA).
• This dynamic approach aims to provide an "instantaneous snapshot" of molecular health, capturing immune system changes that serve as a common denominator for various diseases.
• Unlike static testing, Freenome's application requires serial sequencing over time to monitor biological changes, similar to repeated imaging.

AI-Driven Diagnostic Evolution

• Freenome leverages machine learning to detect specific disease signals within convoluted immune data, theoretically enabling the detection of any condition involving immune changes, not just cancer.
• Unlike traditional diagnostics where performance degrades post-launch due to unaccounted edge cases, Freenome's AI can improve accuracy after launch by learning from real-world results.
• This iterative learning loop creates a potential "clinical trial" scale exceeding 120,000 participants, vastly larger than the largest pre-market trials ever conducted.

Clinical Risks and Misinterpretation

• A specific lawsuit in Oregon involved a patient undergoing a hysterectomy and bilateral mastectomy based on a genetic variant of unknown significance in the MLH1 gene, which has a weak association with breast cancer.
• This case highlights the critical challenge of communicating complex, uncertain genetic data to non-geneticist physicians.
• The speaker notes that standard screening methods like mammography have a 50% false positive rate, suggesting current diagnostics often lead to over-diagnosis or misdiagnosis.

Reimbursement and Economic Barriers

• Currently, only ~20% of diagnostic tests sold receive full reimbursement; the remaining 80% often lack clear payer coverage.
• Payers generally require proof of return on investment (e.g., cost savings from early detection) before covering new tests.
• Emerging payment models include partnerships with life insurance companies and closed hospital systems to bypass traditional payer limitations, though these remain in early stages.