EP 231: From polygenic scores to AI-driven medicine with Andrea Ganna of FIMM

Value of the Information & Strength of the Argument

The value of the information is high, as it provides expert insights into the current state and future directions of using genetics and AI in medicine. Ganna’s arguments are well-structured, distinguishing between prognostic and predictive enrichment, and he supports his claims with references to specific studies and his own research. He acknowledges limitations, such as the lack of good PRS for trial endpoints and the challenges of learning causality from observational data. The discussion is balanced, avoiding overhype, and offers practical considerations for implementing these technologies.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is strong, with Ganna referencing peer-reviewed studies and his own published work. The sources cited in the description are relevant and credible (PMC articles). The title accurately reflects the content, which covers both polygenic scores and AI-driven medicine. The discussion is grounded in evidence, and Ganna is careful to distinguish between established findings and speculative ideas. The podcast format allows for in-depth exploration, and the host asks probing questions that clarify technical points.

The title accurately reflects the content, which transitions from polygenic scores to AI-driven medicine, covering both topics in depth.

8/10

The discussion is led by an established researcher (Andrea Ganna) with relevant expertise in genetics and biobank-scale data. The claims are grounded in published studies (linked in the description) and the conversation is nuanced, acknowledging limitations. However, as an interview, it relies on expert opinion rather than presenting new primary data.

• 00:00Intro to The Genetics Podcast
• 01:00Welcome to Andrea
• 01:53Andrea’s research focuses, including polygenic scores in biobanks and AI applications
• 03:13Complementarity between polygenic scores and electronic health record–derived risk signals across biobanks
• 05:08Using polygenic risk scores for prognostic versus predictive enrichment in clinical trials
• 11:08Limitations and opportunities of using AI models on large-scale electronic health records
• 16:57Legal, data infrastructure, and privacy barriers to building AI models on health records
• 19:33Choosing model architectures for healthcare AI
• 21:30Using AI and multi-omics data to integrate biological knowledge and the challenge of learning causality
• 23:37How removing genetic effects from proteins improves disease prediction and highlights the role of environment
• 27:02Finland’s health data ecosystem and national biobanks
• 30:46Using genetics to improve trial emulation in biobank data and observational studies
• 36:36Closing remarks

• Trial emulation study — Referenced in the description as a study related to using genetics to improve trial emulation.
• Polygenic scores study — Referenced in the description as a study on polygenic scores.
• Podcast show notes — Link provided in the description for show notes.

• Polygenic Risk Scores and Clinical Utility — This article discusses the clinical utility of polygenic risk scores, aligning with the episode's discussion on their use in trials.
• AI in Healthcare — This paper reviews the application of AI in healthcare, supporting the episode's focus on AI-driven medicine.

• Limitations of Polygenic Risk Scores