Contrary to their name, rare diseases are not a niche issue. Citing a 2019 study, Rabinowitz reframes them as a massive socioeconomic burden, costing the U.S. a trillion dollars annually. This is split between $500 billion in direct medical expenses and $500 billion in lost productivity for families navigating long diagnostic odysseys.
Rabinowitz's entry into diagnostics wasn't driven by academic interest but by two family tragedies: the loss of his sister's child to Down syndrome complications and later, the loss of his own child to a genetic condition. This visceral, personal motivation fueled his relentless pursuit of better prenatal testing technology.
In a battle of methods, Natera's deep learning AI, trained on millions of samples classified by classical statistical models, began to outperform its teachers. The AI was better at identifying the underlying noise and difficult outlier cases, demonstrating a non-obvious capability of AI to find patterns beyond its explicit training logic.
Despite Natera's test for 22q11 microdeletions showing high efficacy and getting backing from medical genetics societies, it still lacks broad insurance reimbursement and key guideline approval. This socioeconomic bottleneck means hundreds of families suffer each year, highlighting that technology often outpaces the adoption infrastructure.
Rabinowitz recounts a pivotal conversation with Paul Barron, the inventor of packet switching. Barron framed biology's fundamental components as "the transistors of biology," suggesting that while the silicon revolution's impact on quality of life might be plateauing, the biological frontier was just beginning, offering a chance for world-changing impact.
Myome and Natera are building foundational models for oncology that function like genomic language models. By training on vast cancer sequence and clinical data, these models learn the context of a patient's disease to predict the next mutation, similar to how transformers like GPT predict the next word in a sentence.
Matthew Rabinowitz provides a powerful economic metric for innovation in diagnostics. He states that for every single percentage point of increased sensitivity at a fixed specificity achieved by genetic and AI models, the U.S. healthcare system saves approximately $7 billion in direct medical costs. This makes iterative improvement a massive economic imperative.
Rabinowitz shares how his team, working on predicting HIV drug resistance in 2005, found that neural networks underperformed convex methods like support vector machines. They concluded complex problems required constrained models, completely missing the future potential of large-scale data and stochastic methods that would later empower deep learning, a key lesson in technological humility.